The Cognitive Neuroscience of Memory: An Introduction

Citation preview

The Cognitive Neuroscience of Memory

This page intentionally left blank

THE

COGNITIVE NEUROSCIENCE OF MEMORY an introduction

Howard Eichenbaum Laboratory of Cognitive Neurobiology Boston University Boston, Massachusetts

OXFORD UNIVERSITY PRESS

2002

OXFORD UNIVERSITY PRESS

Oxford New York Athens Auckland Bangkok Bogota Buenos Aires Cape Town Chennai Dar es Salaam Delhi Florence Hong Kong Istanbul Karachi Kolkata Kuala Lumpur Madrid Melbourne Mexico City Mumbai Nairobi Paris Sao Paulo Shanghai Singapore Taipei Tokyo Toronto Warsaw and associated companies in Berlin Ibadan

Copyright © 2002 by Oxford University Press, Inc. Published by Oxford University Press, Inc. 198 Madison Avenue, New York, New York 10016 http://www.oup-usa.org 1-800-334-4249 Oxford is a registered trademark of Oxford University Press. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior permission of Oxford University Press.

WL

Library of Congress Cataloging-in-Publication Data Eichenbaum, Howard. The cognitive neuroscience of memory : an introduction/ Howard Eichenbaum. p.;cm Includes bibliographical references and index. ISBN 0-19-514174-1 (cloth)—ISBN 0-19-514175-X (pbk.) 1. Memory. I. Title [DNLM: 1. Memory—physiology. 2. Brain—physiology. 3. Cognition—physiology. 102 E338c 2002] QP406 .E334 2002 612.89—dc21 2001036607

135798642 Printed in the United States of America on acid-free paper.

Preface

This book is written for undergraduate students and others who seek an overview of progress in understanding how the brain accomplishes one of its most marvelous acts, memory. At the outset I review the history of thinking and research on the biological bases of memory, and highlight discoveries made in a "Golden Era" that spanned from the late nineteenth century into the twentieth century. During this period major breakthroughs were made, revealing secrets about the fundamental elements of the brain and how they work. Although these discoveries were about brain function in general, many of the researchers were interested in the applicability of their findings to the phenomenon of memory. Also, during the Golden Era four main themes in memory research were initiated. In my introduction, I attempt to give the reader an appreciation for how those themes emerged from the discoveries made in that period. Those four themes provide the framework for the remainder of the book. The first theme is "connection," and it considers how memory is fundamentally based on alterations in the connectivity of neurons. This section of the book covers the most well-studied models of cellular mechanisms of neural plasticity that may underlie memory. The second theme is "cognition," which involves fundamental issues in the psychological structure of memory. This section of the book considers the competition among views on the nature of cognitive processes that underlie memory, and tells how the controversy was eventually resolved. The third theme is "compartmentalization," which is akin to the classic problem of memory localization. However, unlike localization, the notion of "compartments" is intended to avoid the notion that particular memories are pigeonholed into specific loci, and instead emphasize that different forms of memory are accomplished by distinct modules or brain systems. This section of the book surveys the evidence for multiple memory systems, and outlines how they are mediated by different brain structures and systems. The fourth and final theme is "consolidation," the process by which memories are V

vi

Preface

transformed from a labile trace into a permanent store. In this section of the book, I summarize our current understanding of two distinct stages in memory consolidation. One stage involves molecular and cellular mechanisms that underlie a fixation of changes in the connection strengths introduced earlier. The other stage of consolidation occurs at the level of brain systems and involves a reorganization and restructuring of the circuits that store and retrieve memories. For heuristic purposes I attempt to deal with these two stages separately, although they are highly interrelated by the brain mechanisms involved. I hope the book will be of use to cognitive scientists, biologists, and psychologists who seek an introduction to biological investigations of memory, and to undergraduate students seeking an expanded coverage of the neurobiology of memory for courses in learning and memory or behavioral and cognitive neuroscience. Readers will benefit from a solid background in basic molecular biology and neurobiology. However, a brief overview of the necessary biological background is included. It bears mentioning that substantial portions of the material presented in this book are derived from another recent book, From Conditioning to Conscious Recollection: Multiple Memory Systems of the Brain (Oxford University Press), coauthored by myself and Neal Cohen. Although there is overlap in the materials of the two books, they differ substantially in two ways. First, the earlier book has a much more limited scope. It is a comprehensive presentation and synthesis, as well as an attempt at reconciliation of current controversies, on the specific topic of multiple memory systems. Its aim is focused on a thorough analysis of one of the central themes of the present book, the theme of "compartmentalization." By contrast, the scope of the present book is much broader. It provides a general introduction to the history of brain and memory research, and is constructed as a comprehensive survey of topics in memory research, including the molecular and cellular bases of memory in cells, invertebrate model systems, and vertebrate systems, the psychological foundations of learning theory, and the phenomena of memory consolidation, as well as the topic of multiple memory systems. Second, the earlier book is constructed for an advanced readership, primarily scientific colleagues and graduate students with considerable previous background in relevant areas of neuroscience and the neurobiology of memory. By contrast, the treatment of topics in the present book is introductory. This book intentionally makes no effort at being detailed or thorough in the described experiments in any research area. Rather, it is aimed to show how we address central questions in brain mechanisms of memory, and seeks to provide a basic un-

Preface

vii

derstanding of each of several central issues through a presentation of a selected set of classic and recent exemplary experiments. In addition, for students and instructors who plan to use this book as a text, I have included a set of heuristic aids. First, I emphasize a historical perspective at the outset, with a review of memory research from its very beginnings. Second, the book is divided into four sections, each of which distinguishes a fundamental theme in memory research. Each section is introduced with a theoretical and historical overview. Third, each chapter begins with a set of "Study Questions" aimed to guide the student toward the central issues in that chapter. I encourage students to think about these questions as they read the text, and then write out detailed answers in preparation for exams. Fourth, each chapter ends with a "Summing Up" section that recaps the major take home points in that chapter and attempts to synthesize the several issues that arose. Fifth, I have included a Glossary that contains definitions of frequently used and important terms. In addition, it may be useful to have a basic neuroscience text available as an adjunct reference for the course. Such a text will help orient students to anatomical terms and provide supplementary information about the anatomy and physiology of brain structures described here. It is hoped that these aids will help students in formulating their "schemas" for the topics in this book and permit them to take more away from the information in it. Considerable appreciation goes to Michelle Barbera who created the illustrations for this book, and to Fiona Stevens of Oxford University Press for her counsel on its organization and content. Thanks also to Neal Cohen, my long time collaborator in thinking and writing about memory, including the related co-authored book described above. More generally, I am indebted to the many students whom I have had the pleasure of teaching over the last couple of decades; they have contributed by asking the hard questions and demanding clear explanations, some of which I hope are conveyed in this text. Finally, I owe a fundamental debt to Edith Eichenbaum, a professional teacher like me, who has provided generous encouragement and guidance throughout my scholarly life.

This page intentionally left blank

Contents

1. Introduction: Four Themes in Research on the Neurobiology of Memory, 1 Part I Connection: The Cellular and Molecular Bases of Memory, 27 2. Neurons and Simple Memory Circuits, 29 3. Cellular Mechanisms of Memory: Complex Circuits, 53 Part II Cognition: Is There a "Cognitive" Basis for Memory? 79 4. Amnesia—Learning about Memory from Memory Loss, 85 5. Exploring Declarative Memory Using Animal Models, 105 6. Windows into the Workings of Memory, 139 Part III Compartmentalization: Cortical Modules and Multiple Memory Systems, 171 7. The Cerebral Cortex and Memory, 175 8. Multiple Memory Systems in the Brain, 195 9. A Brain System for Declarative Memory, 213 10. A Brain System for Procedural Memory, 237 11. A Brain System for Emotional Memory, 261 ix

x

Contents

Part IV Consolidation: The Fixation and Reorganization of Memories, 283 12. Two Distinct Stages of Memory Consolidation, 285 13. Working with Memory, 311 Final Thoughts, 339 Glossary, 341 References in Figure Captions, 353 Index, 357

The Cognitive Neuroscience of Memory

This page intentionally left blank

I

Introduction: Four Themes in Research on the Neurobiology of Memory

STUDY QUESTIONS What is the neurobiology of memory? What major questions about memory are pursued with a neurobiological approach, and how are these questions addressed in experimental analyses? What are meant by "connection," "cognition," "compartmentalization," and "consolidation "?

o

ur memories reflect the accumulation of a lifetime of experience and, in this sense, our memories are who we are. Surely the background of our makeup is determined largely by our genes; genetics sets the range of what we can aspire to be. However, by contrast to generality of genetic limitations, the specifics are a matter of memory. We learn to walk, to dance, to drive a car, to throw a ball, and to play a video game—a myriad of acquired skills we come to take for granted. We learn to fear dangerous situations, to appreciate particular types of music and styles of art— a broad range of aversions and enjoyments we have assumed as elements of our preferences and personality. We learn to speak, and to speak and understand our particular language. We learn world history, and we learn our own family tree and personal autobiography—all of these, and much, much more, compose the vast contents and intricate, complex organization of memories that make each of us a unique human being. So, the

1

2

Introduction

analysis of memory is a search for self-understanding, an adventure that promises to reveal the inner secrets of how we came to be who we are. The nature of memory—its basic biological structure, its psychological character and organization, and its longevity—has been the subject of investigations by philosophers, writers, and scientists for hundreds of years, and each approach offers its own distinct avenue for understanding memory. Recently, with the rise of modern methods of cognitive science and neuroscience, and their combination, many new and deep insights about the mechanisms of memory have emerged. These observations have also led us to a greater general understanding about the mind and about brain functions that mediate cognition, emotion, behavior, and consciousness. The aim of this book is to explore memory from the perspective of cognitive neuroscience, offering a historical and a current overview of how brain functions in memory have been studied and what we have learned about memory as an encompassing aspect of the mind. The present chapter introduces some of the philosophical and historical underpinnings of research on the biological bases of memory. I begin by presenting four central themes that have guided memory research for over a hundred years. Substantial preliminary evidence regarding each of these themes emerged during a "Golden Era" for neuroscience in the latter half of the nineteenth century and the beginning of the twentieth century. A brief introduction to some of these accomplishments provides the background for a subsequent, more detailed summary of progress on each of the four central themes in the remainder of the book. The four "C's" At the outset of systematic investigations of the nervous system four major themes dominated considerations of brain function of relevance to memory. In part to facilitate your memory for them, I refer to these themes as the four "C's": connection, cognition, compartmentalization, and consolidation. The first theme—connection—concerns the most basic level of analysis of memory function, the basic nature of the circuitry of the brain including the elements of information processing and how they communicate with one another in the service of memory. The emphasis on "connection" here reflects the major conclusion that has emerged in this research—that memory is encoded within the dynamics, that is, the changeability or plasticity, of connections between nerve cells. More specifically, the consensual view from the perspective of memory as a phenomenon of brain cells is that memories are instantiated by alterations of the strength

Four Themes in Research on the Neurobiology of Memory

3

or reliability of communication between cells via their synaptic connections. Efforts to understand the nature of the storage mechanism are aimed at identifying the biological materials, the cellular "switch," from which memories are made. Accordingly, cellular and molecular analyses of memory have been focused on characterizing the modifiability or plasticity of neural connections that underlie memory performance. Experimental research in this area seeks to discover specific molecular events that form the basis of a memory, and subcellular structures that support neural connections that accomplish memory. This research also asks how the basic memory storage mechanism can be modulated by natural neurochemical events and manipulated by genetic and pharmacological interventions. The second theme—cognition—refers to the nature of memories at the highest level of analysis, the psychological level. Consider the following unusual instance of memory. You walk into a new room, an odd colored light fills the room, then a loud and frightening rattling noise goes off and persists—you exit with haste. Subsequently, you happen to enter that room again, and again the odd light comes on. You immediately leave. Is the memory for that event represented in terms of a new association between particular novel sensory stimulus, the odd light, and the escape behavior you executed in response to it, what psychologists call a "stimulusresponse association"? Or is the fundamental association between the light and the consequent irritation the loud sound evoked in you, what psychologists call a "stimulus-reinforcer association"? Or are the concepts of stimulus-response and stimulus-reinforcer associations altogether too simplistic a view of how your memory is stored? Does your memory representation contain a record of the entire series of relevant and peripheral events that constituted the learning episode? Is that memory isolated among a large and loose collection of episodic memories? Or is that memory contained within a systematic organization of other experiences in the same building that form a network of knowledge about your experiences there? Questions about the nature of memory representations have been at the center of debates about whether the complexity of memory can or cannot be reduced to a set of simple associative principles. During the Golden Era there were divergent and strong views espousing either that memory can be simplified to a set of principles about stimulus-response and stimulus-reinforcer associations or that memory involves complex networks that can be understood only in terms of cognitive operations that are not reducible to simple associative mechanisms. The understanding that has emerged from this area of research is that there are mechanisms that guide behavior from stimulus-response and from stimulus-reinforcer associations, and there is a "cognitive" form of memory that is distinguished in

4

Introduction

both its psychological mechanisms and anatomical pathways from the other forms of memory. The third theme—compartmentalization—addresses the question of memory localization. This question appeared early on in debates about whether memory can be localized to a particular area of the brain, especially the cerebral cortex, or whether it is distributed throughout the cortex, or indeed the entire brain. The major conclusion that has emerged on this theme is that memory as a whole is distributed widely in the brain. But, at the same time, different kinds of memory are accomplished by specific brain modules, circuits, pathways, or systems. That is, memory is compartmentalized. The compartmentalization occurs at two levels. First, the cerebral cortex is composed of many anatomically circumscribed "modules," each of which makes a correspondingly specific contribution to memory function. Second, there are multiple memory systems in the brain, all of which involve the cerebral cortex but they diverge in pathways leading from the cortex to other structures that lie beneath the cortex (subcortical structures), and the systems formed by these cortical-subcortical pathways accomplish different kinds of memory. This research area also considers the relationship between memory and other cognitive processes, including consciousness, coordinated movement, and emotion. The major aim of research on brain modules and systems is to identify and distinguish the different roles of specific brain structures and pathways, usually by contrasting the effects of selective damage to specific brain areas. Another major strategy in attacking this issue focuses on localizing brain areas that are activated, that is, whose neurons are "turned on" during particular aspects of memory processing. Some of these studies use new functional imaging techniques to view activation of brain areas in humans performing memory tests. Another approach seeks to characterize the "code" for memory within the activity patterns of single nerve cells in animals, by asking how information is represented by the activity patterns within the circuits of different structures in the relevant brain systems. The fourth theme—consolidation—concerns when and how memories become permanent. It is well known that some experiences are rapidly forgotten, whereas others are remembered for a lifetime. And there have been many anecdotal and clinical reports that various forms of interference or head or brain injury can "wipe out" memories that were recently acquired but have less effect on memories acquired remotely before the interfering event or injury. These observations suggest that memories are initially labile and later become resistant to loss, suggesting a process of consolidation during which memories take on a permanent form. Mod-

Four Themes in Research on the Neurobiology of Memory

5

ern research has shown that there are two general kinds of consolidation. One of these—which I call "fixation"—involves a cascade of molecular and cellular events during which the changes in connections between cells become permanent in several minutes to hours after a memory is formed. This process can be influenced by many factors, including among them a specific brain system for modulation of memory fixation. The other kind of consolidation process is called "reorganization" because this process involves a prolonged period during which distinct brain structures interact with one another, and the outcome is that newly acquired information is integrated into one's previously existing body of knowledge. This reorganizational process therefore involves an entire brain system, and discovering how it works involves a consideration of both the individual contributions of particular parts of the brain system and the nature of interactions among the parts.

The Golden Era The aim of this chapter is to set the stage for the succeeding sections that will review our understanding of each of the four major themes in memory research, and in doing so provide a framework for understanding the neurobiological bases of memory. I pursue an historical approach, elaborating on each of the four "Cs," beginning with a summary of discoveries made at the threshold of modern neuroscience research on memory in the latter half of the 1800s. Before that period, some of the critical background had already long been established. In 1664 the early anatomist Willis published the first description of the anatomy of the brain and had suggested that different brain areas controlled distinct functions. Simplified views of the brain and its main components are provided in Figures 1-1 and 1-2. Also, in 1791 Galvani introduced the notion that electricity is the mechanism of nervous conduction. But around the turn of the twentieth century several additional major discoveries formed the full beginning of a scientific analysis of the brain, a Golden Era in which several key findings led toward real progress in understanding brain function and memory. Some of these contributions represented major advances to the field of neuroscience in general, and others pertained to memory research in particular. Here I highlight a few of the major insights of that period that have had lasting impact. Put together, these observations should give the reader a sense of the field of neuroscience, and especially about the neurobiology of memory, upon which all subsequent progress is based.

6

Introduction

Figure 1-1. Side view of the human brain, showing the major subdivisions of the cerebral cortex, the cerebellum, and terminology used to describe relative locations in the brain.

Connection: Cellular substrates of brain communication and memory One main set of advances that laid the foundation for memory research involved discoveries about the basic building blocks of brain circuits. These discoveries identified the fundamental elements of the brain, characterized

figure 1—2. Cut-away side view schematic of a generic mammalian brain and spinal cord, showing locations of major subcortical brain structures and the system of ventricles (a hollow circulatory system running throughout).

Four Themes in Research on the Neurobiology of Memory

7

how they are assembled into simple functional circuits, and demonstrated that they can be modified during learning. The neuron doctrine One major area of discovery about the brain that contributed directly to our understanding of the cellular and molecular substrates of memory was the development of the "neuron doctrine" by the Spanish anatomist Santiago Ramon y Cajal. The neuron doctrine is the notion that the brain is composed of discrete nerve cells, and that these cells are the essential units of information processing, connected to one another so as to transfer and integrate information in large-scale networks. At the time of Cajal, this view was not entirely new, but it was also not widely accepted. Rather, Cajal's work addressed a major controversy about the nature of connections between neurons. In the debate, one camp, called the "reticularists," argued that the brain is a single unified interconnected network of fibers in which all the cells were fused to one another. By contrast, the other camp, called the "antireticularists," suspected that the brain was composed of independent nerve cells as units, but they had no definitive evidence. Before Cajal, the strongest argument for independent cells came from the observation that small lesions in one area resulted in sharply defined areas of degeneration, not what one would expect of a fused network. Cajal's success was based on his adoption of a new staining method that was developed by another anatomist named Camillo Golgi in 1873. The method involved a "black reaction," a new silver stain that had the remarkable quality of darkening the entire cell membrane of a neuron. At the same time, the staining was selective to only a small fraction of the neuron population in an area of brain tissue. Thick sections of the brain stained this way provided a full view of individual cells standing out clearly against the background of many other surrounding pale cells. Using this method Cajal was able to provide the most striking confirmation of the already existing identification of the major elements of nerve cells. As shown in Figure 1-3, these include the cell body, the multiple fine processes that extended from one end of the cell body called dendrites, and the single larger process extending from the other end of the cell body called the axon. Also, he noted the specialization of the axon as it contacted the dendrites of other cells; this specialization would later be called the synapse (see next section). Cajal attempted many variations of the procedure. He found that tissue with less myelin, the insulation layer of axons, produced the clearest images of neuronal processes, and the best cases were found in young brains and in birds. He also found that thicker sections allowed one to examine all of the extensions of the cell membrane that connect with other

8

Introduction

Figure 1-3. Drawing by Cajal based on a section through the cortex of a 20-day-old mouse. Note the different types of cells, all oriented vertically from the superficial layer at the top to the corpus callosum (indicated by the letter A) at the bottom, where the axons of the cortex bundle (from DeFelipe and Jones, 1988).

cells. In each of several different preparations he found elaborate endings of axons in nests or baskets of axonal "arborization"—a treelike branching—of the axon connecting it to multiple parts of another cell or multiple cells. In no case did he observe the stain continuing into the next cell, as would be expected if there was a fusion of the axonal ending with dendrites of another cell. Cajal concluded that there must be some method of communication between cells that did not involve a joining of their membranes. Cajal's preparations and the evidence they provided were elegant, and convinced other anatomists and physiologists that each cell was contained within a membrane and was separate although in contact with other cells. These observations won him the Nobel Prize in 1906. Cajal was also able to make key conclusions about the function of neurons from his observations on their anatomy. As said previously, he confirmed the existence of all of the major components of the nerve cell. Moreover, in his studies on visual and olfactory sensory structures, Cajal noted that the dendrites pointed toward the outside world and that the axon pointed toward the brain. From these observations he deduced that nerve cells were functionally polarized, such that information flows from the dendrites to the axon and is subsequently conveyed by the specialized connection to the dendrites of another nerve cell. These conclusions established the essential view that the integration of information occurred by the summation of signals converging from the axons of several neurons onto the dendrites of cells receiving those inputs. In addition, Cajal developed some important and prescient ideas directly

Four Themes in Research on the Neurobiology of Memory

9

relevant to the basic memory mechanism. He studied the brains of several species and observed that the vertebrates higher in the phylogenetic scale had a greater number of connections between nerve cells. He concluded that the increase in connectivity could be the basis of greater intellectual power of the higher species. He suggested that mental exercise could facilitate increased connectivity through a greater number and intensity of the connections, and that these changes in connectivity could coincide with the acquisition of skills such as playing a musical instrument. Cajal's insights have become key axioms for the study of neuronal function and communication. Moreover, his suggestion that "plasticity" in number and strength of neuronal connections underlies learning guides the search for molecular and cellular substrates of memory. The reflex arc

In the same period other major advances were made from studies on the physiology of the nervous system. Perhaps most important among these were Charles Sherrington's observations on the nature of reflexes in the spinal cord. Before Sherrington the existence of involuntary muscle actions was already well recognized, including the basic observation that specific sensory stimulation could be "reflected," as if by a mirror, to generate muscle movements—hence the "reflex arc" (Fig. 1-4). In addition it was known that complex reflexes, such as those mediating jumping in frogs or coordinated flying movements in birds, could be elicited even following decapitation, suggesting control of complex coordination could happen at a level below the brain—at the level of the spinal cord. And it was clear that reflex arcs accomplished within the spinal cord could be inhibited by higher level control. However, there was very little understanding of the underlying circuitry that accomplished either simple or complex reflexes. Sherrington made many contributions that provided the foundations for our understanding of neural circuitry that are as relevant today as they were when he made his discoveries. Even before Cajal's convincing anatomical evidence was provided, Sherrington had reached the conclusion that neurons must be independent elements. Part of the evidence came from his studies on neural degeneration, showing that cortical lesions, damage induced by heat or cutting, resulted in restricted, not diffuse patterns of degeneration. He also realized that the neuron doctrine, and the detailed evidence showing the connection was from axons to dendrites, could explain why neural transmission was one-way. And the discontiguity between cells provided a mechanism for why there was a time lag in reflexes such that they were much slower than predicted from the speed of conduction of the neural impulse— the loss of time in long-range conduction had to involve the extra time re-

10

Introduction

Figure 1-4. A schematic diagram of some components of the pain withdrawal reflex arc. Pain to the skin activates a sensory neuron whose cell body is just outside the spinal cord. The axon synapses with an interneuron within the spinal cord, which excites a motor neuron that causes contraction of the relevant flexor muscle, and excites an inhibitory interneuron that decreases activity in motor neurons that activate the extensor muscles in the same limb.

quired for an impulse to jump a "gap" between the neural elements. Sherrington is credited for inventing the term synapse to describe the hypothetical junction between neurons where transmission occurs. In his classic studies on the "knee jerk" reflex, Sherrington demonstrated the details of his model reflex arc, wherein specific sensory information is gathered at the input end of the arc and then relayed to turn specific muscles on and off at the output end. He provided critical evidence for the existence of the "sixth sense"—specialized sensory receptors in the muscle that monitor muscle length and tension, and showed that these sensory elements send nerves into the spinal cord. He also contributed to the characterization of the maps of sensory inputs from the skin surface into the spinal cord, the so-called dermatomes. By isolating and stimulating sensory roots he could determine which regions of the skin evoked reflexive movements. Conversely, to map the output pattern, he stimulated individual spinal cord motor roots and characterized which muscles were activated. Important as these discoveries were, perhaps Sherrington's most outstanding contribution was elucidating some of the complexities of the cir-

Four Themes in Research on the Neurobiology of Memory

11

cuitry that underlie reflexes. He concluded that, even at the level of the spinal cord, reflex arcs are not entirely separate or independent. In these circuits the initial sensory neurons in the spinal cord receive dedicated inputs, but the output neurons of even the simplest reflexes received information from many of those input cells (Fig. 1-4). Thus, consistent with Cajal's observations on the anatomy of nerve cells, there was a summation or integration of both inputs into a common path that could guide complex behavioral actions. And there were two major kinds of inputs that combined in this integration. There were excitatory inputs by which an impulse from the input cell increased the likelihood of impulse generation in the next cell. And there were inhibitory inputs by which an impulse in the input cell decreased the likelihood of impulse generation in the next cell. He also made major discoveries about the phenomenon of reciprocal innervation, the basic mechanism by which excitatory and inhibitory influences are coordinated to mediate movement. Within this scheme every action by one muscle is coordinated with an opposing action of complementary muscles. For example, the simple act of walking involves the coordinated actions of flexing one group of muscles during the extension of others, followed by the opposite complementary actions in executing the next step in walking. In addition, Sherrington employed the technique of decerebration, cutting the spinal cord just behind the brain, to show that complex coordinated actions existed even without higher level cerebral control. He showed, for example, in a decerebrate cat, when a forelimb was excited to move forward, the hindlimb on the same side moved back and the two legs on the other side of the body exhibited the opposing movements, producing the pattern seen in normal walking, but without conscious cerebral control. In studies on the "scratch reflex" he showed the specificity of arm movements for scratching evoked in response to small areas of skin stimulation. Furthermore, in extensions of these studies he revealed that coordinated and directed scratching movement patterns could be played out over time, with alternating extension and flexion of muscles at different levels of the limb to produce repeated scratching movements, plus postural adjustments in the other limbs to support standing without the use of the scratching limb. Sherrington envisioned all of this as accomplished within the spinal cord, by a "chaining" of reflexes wherein successive coordinated movements are elicited by their predecessors. These elements provided the outline for his formulation on the integrative action of the nervous system. In this prototype for modern views, Sherrington proposed a hierarchy of coordinated control wherein the cerebral cortex is acknowledged as the

12

Introduction

newest and most complicated switchboard of reflexes, and successively lower centers in the brain stem and spinal cord to mediate more and more specific, yet still complex and coordinated actions. The conditioned reflex Initially unrelated to this course of research, the Russian physiologist Ivan Pavlov was making landmark discoveries about the nervous control of digestion. He discovered that the release of digestive fluids was controlled by the nervous system, contrary to the prevailing view that digestive fluids were released as a consequence of mechanical stimulation by food directly onto the stomach wall. To test his hypothesis that the nervous system was involved, Pavlov developed a surgical procedure in which he severed the gullet and attached both open ends to the skin of the neck. This allowed him to either introduce food into the mouth and upper gullet and then retrieve it without going to the stomach, or introduce food directly into the stomach. He found that food stimulation associated with ingestion caused the release of gastric fluids in the stomach, even when the food never reached its normal target. He concluded that food excites the gustatory sensory apparatus in the mouth and gullet, transmitting signals into the brain stem, which, via the vagus nerve, controls the release of gastric fluids. From a neurophysiological perspective, one can say that Pavlov identified a reflex arc for digestion, and for this he received the Nobel Prize in 1904. But by the time he received the prize, Pavlov had already turned his interest to an intriguing report that gastric juices of a horse could begin to flow not only with a direct application of gustatory stimuli but also even when the animal only caught sight of hay. Pavlov replicated the phenomenon of "psychic secretion" using dogs and measured the generation of fluids from the salivary glands. He found that the sight of a piece of beef indeed caused salivation, but he also found that the phenomenon was unreliable. The salivation tended to decrease following repeated presentations of the sight of beef. Conversely, sometimes salivation was initiated by events that preceded the sight of beef, for example, when the person who regularly provided the food merely appeared in the testing room. Pavlov set out to meticulously control the stimuli available to the animal, and he tried out many arbitrary stimuli—including the famous bell rung prior to the presentation of food. Based on the results from a broad range of experimental manipulations, Pavlov concluded there were two kinds of reflexes. One kind of reflex is "unconditioned," identical with the innate and stable reflexes of Sherrington. The unconditioned reflex is composed of a particular unconditioned stimulus (US) that inevitably elic-

Four Themes in Research on the Neurobiology of Memory

13

its its characteristic unconditioned response (UR). The other kind of reflex is "conditioned," that is, acquired through experience. This kind of reflex is the unstable one, and is composed of an arbitrary conditioned stimulus (CS) that when paired with a US comes to elicit a conditioned response (CR) similar in form to the UR. Pavlov identified the critical importance of two parameters in establishing and maintaining the conditioned reflex: the close temporal contiguity of the CR and US—the CS must lead the US by a particular time interval, and the CS must consistently predict the US. Combined, the contributions of Cajal, Sherrington, and Pavlov, as well as many others, laid the basic framework for succeeding views of brain circuitry and function, as well as its role in memory. They showed that the basic elements of the circuits are independent neurons that communicate across synapses, that these elements are integrated within complex patterns of circuitry for coordinated action built up from simple reflex arcs, and that these circuits can be modified to support learned reflexes. These insights set the stage for future investigations on the mechanisms of how connections between neurons are modified during learning, and guided the development of views on the organization of memory for both simple and complex behaviors. Cognition and memory During the same period, distinct developments were made toward characterizing the nature of memory from a purely psychological perspective. Two main and competing lines of theorizing developed. One school, called "behaviorism," developed out of a desire to provide a rigorous science of memory consistent with the findings on the neurophysiology of conditioned reflexes, and attempted to explain all of learned behavior on the basis of elements of association and conditioned responses. The other school, called "cognitivism," emphasized the complexity of learned behavior, and its promoters could not be persuaded that all aspects of cognition, insight, and planning could be captured in an elaborate account of associations or reflex chains and instead required a more elaborate conception and, correspondingly, a more complex neural instantiation. Behaviorism

The tradition of rigorous methodology in memory research began with Herman Ebbinghaus, who admired the mathematical analyses that had been brought to the psychophysics of perception, and he sought to develop similarly precise and quantitative methods for the study of memory. Bas-

14

Introduction

ing his work on a large number of pioneering studies, in 1885 Ebbinghaus published a monograph that set a new standard for the systematic study of memory. Ebbinghaus rejected the use of introspection as a methodology that was prominent in previous conceptual schemes about memory. In its place he developed several key new techniques that would control the nature of the material to be learned and provide quantitative objective assessments of memory performance. To create learning materials that were both simple and homogeneous in content Ebbinghaus invented the "nonsense syllable," a meaningless letter string composed of two consonants with a vowel between (e.g., "ket," "poc." "baf"). With this invention he avoided the confounding influences of what he called "interest," "beauty," and other features that he felt might affect the memorability of real words. In addition, the nonsense syllable simultaneously equalized the length and meaningfulness of the items, albeit by minimizing the former and eliminating the latter. Furthermore, to measure memory Ebbinghaus invented the use of "savings" scores that measured retention in terms of the reduction in trials required to relearn material. In addition, he was the first to employ mathematical-statistical analyses to test the reliability of his findings. It was also in this period that systematic studies on animal learning and memory had their beginnings. In 1901 Small introduced the maze to studies of animal learning, inspired by the famous garden maze at Hampton Court in London (Fig. 1-5). He began what would become an industry of systematic and quantitative studies to identify the minute details of how rats acquired specific responses in repetitions of turns taken in the maze. But he observed that within a trial or two rats prefer a shortcut over the response route that had been reinforced on many previous trials, leading Small to conclude that future experiments should investigate the natural biological character of the animal if one is to be able to interpret the findings. These initial observations set forth a major controversy in the field of animal learning. Can learning be reduced to a set of arbitrary associations between external stimuli and behavioral responses, or must one consider issues such as cognition, insight, and motive? At the turn of the century, Edward Thorndike had invented a "puzzle box" in which he observed cats learning to manipulate a door latch to allow escape from a holding chamber. Based on his observations he proposed the "law of effect," which stated that rewards reinforced repetitions of the specific behaviors that preceded them. (This simple law would be reinvented and extensively elaborated by B.F. Skinner in the 1950s to explain all of learned behavior.) In the same period John Watson published his accounts on maze learning by rats. In one of his most famous experi-

Four Themes in Research on the Neurobiology of Memory

15

Figure 1—5. Small's maze, based on the maze at Hampton Court. The animal entered at the bottom of the maze and had to find its way to the central open area.

ments, Watson trained rats to run a maze and then searched for the underlying stimulus control by eliminating one sense after another. He found the rats could still run the maze with only the kinesthetic (muscle) sense remaining, leading him to conclude that the learning must be mediated by a chain of reflexes, consistent with the evidence from physiology. By 1913 he had accumulated sufficiently compelling evidence for the reductionist strategy that he wrote a "behaviorist manifesto," formalizing the view that learning could be understood in terms of simple stimulus and response associations without resorting to considerations of vague concepts such as consciousness. Cognitivism

William James captured prevalent views of the time on the origins of both the behaviorist and cognitivist perspectives in his classic The Principles of Psychology. Within this treatise, James considered reflex mechanisms and pathways as the essential building blocks of memory, and he called these the mechanisms for the formation of a "habit." James viewed habits as built upon a very primitive mechanism that is common among biological

16

Introduction

systems and due to an inherent plasticity of organic materials. Within the nervous system he viewed the mechanism of habit in terms of its known electrical activity, and suggested that electrical currents should more readily traverse paths previously taken. Thus, James felt that a simple habit was nothing more than the discharge of a well-worn reflex path, entirely consistent with the views of emerging behaviorism. Furthermore, James expanded on this notion, attributing great importance to habits as the building blocks of more complicated behavioral repertoires. He suggested that well-practiced behaviors and skills, including walking, writing, fencing, and singing, are mediated by concatenated discharges in connected reflex paths, organized to awaken each other in succession to mediate the serial production of learned movement sequences. However, while acknowledging the importance of habits as the fundamental mechanism that underlies memory, James recognized real "memory" as something altogether different from habit. James argued that there were two forms of memory, differentiated by their timing and by their role. He suggested that initially there is a primary memory, what we today call short-term or working memory, a short-lived state where new information has achieved consciousness and belongs to our stream of thought. Primary memory also serves as the gateway by which material would enter secondary memory, what we now call long-term memory. James emphasized that secondary memory involves both the intellectual content of information we have learned and the additional consciousness of the experience during learning. In addition to the feature of personal consciousness, the full characterization of memory was framed in terms of its structure as an elaborate network of associations. James argued that, while memory is based on the habit mechanism, it is vastly elaborated such that the formation of associations among habits supports the richness of our experience of a memory. Thus, the underlying foundation of recall involves a complex, yet systematic set of associations between any particular item and many other co-occurring items during one's experiences. It is of interest that James also offered speculations that touched on the biological basis of memory. He suggested that memory depends on two aspects of the habit mechanism. First, how good a memory is depends on the strength or persistence of the pathway—this aspect James suggested was innate. Second, he suggested that memory depends on the number of pathways through which an item is associated. He emphasized the latter as more malleable, and argued that the key to a good memory is to build diverse and multiple associations with one's experiences, weaving information and experiences into systematic relations with each other. The ca-

Four Themes in Research on the Neurobiology of Memory

17

pacity to search through one's network associations was held to be the basis of conscious recollection, and could lead to creative use of memory to address new problems. Conversely, James admonished his students not to simply rehearse learned materials. This, he argued, could lead only to the concatenation of habit pathways that could only be expressed by repetition. James never contrasted "habit" and "memory" as distinct forms of memory. A more direct recognition of two forms of memory can be attributed to the philosopher Henri Bergson, who in 1911 explicitly proposed that representations of the past survive under two distinct forms, one in the ability to facilitate repetition of specific actions, and the other in independent recollections. The suggestion that habits and memories might both have a common cellular basis, and at the same time exist as distinct forms of memory, would ultimately resolve the controversy between the behaviorist and cognitivist schools. In addition, the notion of different forms of memory would reappear in the solution to the controversy about the compartmentalization of memory discussed next. Compartmentalization of cortical function and memory In this same period as advances were made in characterizing reflex circuitry, and the controversy about the nature of memory was brewing, a separate battle was engaged over another critical puzzle about the brain and memory. This controversy focused on the organization of the cerebral cortex. There was already scattered evidence from studies of patients with circumscribed brain damage that the anterior (front) and posterior (back) regions of the cerebral cortex played different functional roles. But no clear functional specifications arose from the early observations. Two dramatically different views emerged during the nineteenth century. "Orgonology"

The earliest specific and systematic formulation on cortical localization came in the early 1800s from the German physician Franz Joseph Gall. Following a trend early in the eighteenth century in which scientists were attempting to associate body features with aspects of personality, Gall sought to determine whether there existed variation in structure and function of the brain. He developed a theory of cortical localization, which he called organology, in which each of many independent psychological faculties is mediated by a specialized organ in the brain. The central axiom of this theory was that individual differences in specific faculties were reflected in greater development of the mediating brain organ and, corre-

18

Introduction

spondingly, the size of the overlying skull area. The theory was developed using a combination of observations on individual variation in specific psychological faculties and skull areas in humans and on comparisons between the abilities and skulls of animals versus humans. Gall's detailed investigations on humans included a variety of individuals that represented extreme variations in behavioral capacities. He sought out people with a special talent, such as writers, statesmen, and musical and mathematical prodigies, or with a behavioral abnormality, such as lunatics, the feebleminded, and criminals. For each he would interview the person extensively to characterize their unusual behavioral qualities and carefully examine the head for irregularities. Based on his insights about the functional aspects of their abilities and on discovery of unusual skull features, he envisioned a tight correlation between the skull and brain anatomy and a direct link to the unusual aspect of behavior. For example, among his earliest findings on humans was the observation that some people who were outstanding in memorizing verbal material had bulging eyes, suggesting to Gall an enhanced development of an organ in the frontal lobes specialized for verbal memory. In addition, Gall collected hundreds of skulls from animals and made detailed comparisons between the anatomical features of those skulls and those of humans. The same sort of loose correlation was applied, in this case comparing the psychological abilities of animals both between different species and with humans. From a combination of all this material he devised a system of faculties, some shared by humans and animals and some exclusively human (Fig. 1-6). An example of his reasoning comes

Figure 1-6. Gall's organology scheme. The assignments were: 1. Instinct for reproduction. 2. Love of offspring. 3. Affection. 4. Instinct of self-defense. 5. Carnivorous instinct, tendency to murder. 6. Guile. 7. The feeling of property, theft, hoarding. 8. Pride. 9. Vanity, ambition. 10. Forethought. 11. Educability. 12. Places. 13. Memory of people. 14. Words. 15. Language and speech. 16. Colors. 17. Sounds, music. 18. Sense of connections between numbers. 19. Mechanics of construction. 20. Wisdom. 21. Metaphysics. 22. Satire. 23. Poetry. 24. Kindness. 25. Ability to imitate. 26. Religion. 27. Perseverance.

Four Themes in Research on the Neurobiology of Memory

19

from his deductions about the faculty of "destructiveness, carnivorous instinct, or tendency to murder," localized to an area above the ears. This designation was strongly based on a combination of observations on animals and humans. That area was larger in carnivores than in grasseating animals. And the area was overly large in a successful businessman who gave up his profession to become a butcher, in a student who was fond of torturing animals and became a surgeon, and in a pharmacist who became an executioner. Gall's early attempts to make functional assignments of cortical areas were considerably off-base, due to two major flaws in his methodology. First, Gall's methods were based on individual cases, each of which was subject to considerable interpretation about the nature of the basis of their unusual abilities. Second, Gall simply assumed a close correlation between skull and brain anatomy, and had little interest in examining brain directly. Gall's assumptions led him in the wrong direction in almost every case. Later clinical and experimental studies consistently failed to confirm Gall's specific functional assignments. However, these later studies would demonstrate localization of cortical functions, albeit with different functional designations. Thus, quite rightfully Gall is given considerable credit for the basic insight that the cortex is composed of multiple, functionally distinct areas. Case studies of human patients with localized cortical damage

More compelling evidence for specific functional designations within the cerebral cortex came from observations of neurological case studies of patients with selective brain damage, and from parallel physiological studies. Among the most important of the case studies on brain pathology was one made by the French physician Paul Broca in 1861. The study involved a 51-year-old man named Lebourgne who had suffered from epilepsy since birth and had later lost the power to speak and developed a right side paralysis and loss of sensitivity. This patient came to Broca's attention when he was admitted to Broca's surgical ward at the hospital for an unrelated disorder, and died within a week. The autopsy revealed a circumscribed area of damage in the third convolution of the frontal lobe on the left side of the cerebral cortex. The patient's disorder was also well circumscribed. He was virtually unable to speak—indeed, he acquired the nickname "Tan" from the only sound he made—but his mouth was not paralyzed and he retained the capacity to hear and understand speech. This case provided a compelling demonstration of a highly severe and selective behavioral disorder related to a specific cortical zone. The argument for localization of higher functions was made all the more compelling with the description of a complementary case by Carl Wernike in 1894. In this

20

Introduction

case the patient was severely impaired in speech comprehension, without hearing loss or a disorder of speech production, and the damage was circumscribed to a zone within the left temporal cortex. Experimental neurology and neurophysiology

The evidence from neurological cases was strongly supported by concurrent findings from experimental work on animals. The earliest studies on animals, specifically aimed to test Gall's theory, were reported in 1824 by Flourens. He failed to find localization of sensory and motor functions following cortical damage in birds. These findings and other studies that could not demonstrate selective losses in mammals became the strongest evidence against localization of cortical function. However, the use of birds and other animals with relatively less differentiated cortical areas turned out to be a poor choice for an experimental model in analyses of cortical function. The case for localization was eventually made, from studies using brain stimulation in dogs, by Gustav Fritsch and Eduard Hitzig published in 1870, and with the careful work of David Ferrier using brain lesions in monkeys, presented in 1874. Based on earlier work showing that stimulation of the head or cortex could produce twitching movements of the musculature, Fritsch and Hitzig employed minimal levels of electrical stimulation to map the cortex of dogs. They found that stimulation of a zone within the frontal cortex resulted in specific muscle movements. Moreover, they discovered that minimal stimulation of one cortical area more anterior and dorsal produced selective movements of the forepaw on the contralateral side of the body, whereas nearby regions of stimulation resulted in muscle movements in adjacent body areas. Low-level stimulation of other cortical areas did not produce movements in any part of the body, indicating that they had isolated a specialized motor area of the cortex and that area was organized as a kind of mapping of the musculature of the body. Based on the physiological findings of Fritsch and Hitzig, Ferrier was convinced there had to be separate cortical areas that mediated specific sensory functions, such as vision, hearing, smell, and touch, and other areas that controlled movement. He suspected that previous studies had failed to find selective behavioral-anatomical correlations because the lesions were too small or the animals selected did not have sufficiently differentiated cortical areas. He prepared two monkeys, and in each showed a selective disorder associated with a specific area of cortical damage. One monkey had a severe paralysis on the right side of the body associated with a circumscribed lesion within the left frontal area. The other monkey was completely deaf following a bilateral removal of the temporal

Four Themes in Research on the Neurobiology of Memory

21

lobe. Ferrier's evidence held up to close scrutiny by his colleagues and provided the most compelling initial evidence of distinct sensory and motor functional areas in the cerebral cortex. This combination of studies settled the debate on localization, making it clear that the cerebral cortex does not operate as a unitary organ, but rather is composed of many functionally distinct compartments. Subsequently, the localization controversy would arise again, this time about the locus of memory traces per se, as distinct from the more general issue of functional localization. This time, the strong localizationist view would not hold, at least with regard to the cortex. However, a new perspective, based on anatomical considerations beyond the cortex, would show that memory is subdivided according to larger pathways or compartments that involve connections between the cortex and other brain regions. Consolidation A major topic in research of the Golden Era directly associated with the phenomenon of memory involved studies on memory performance in neurological patients with memory disorders, as well as in normal human subjects. Two phenomena of amnesia following brain damage were prominent. First, patients with memory deficits could acquire new information and remember it briefly, but showed an abnormally rapid amount and rate of forgetfulness. This phenomenon in amnesia, called anterograde amnesia, was intimately tied to the diagnosis of memory impairment, in that the disorder of memory could be contrasted with intact perception and comprehension, as well as a spared ability to hold information long enough to demonstrate the latter capacities. Second, and even more impressive, was the observation of retrograde amnesia, the loss of memories acquired before the brain trauma. Both phenomena of memory loss were systematically studied first in the Golden Era. The neuropothology of memory In 1882, the French philosopher and psychologist Theodore Ribot reviewed a large number of cases of retrograde amnesia associated with brain damage and head trauma. He observed that in those cases where memory impairment is the major consequence, memories acquired remotely before the insult were relatively preserved compared to those acquired recently just before. His formulation, which came to be known as Ribot's law, was stated as a "law of regression" by which the loss of memory is inversely related to the time elapsed between the event to be remembered and the

22

Introduction

injury. Ribot thus concluded that memories required a certain amount of time to be organized and fixed. Further early systematic characterizations of memory disorders, and the incumbent insights they provide about normal memory, began with the descriptions of two forms of dementia in which memory loss plays a prominent role. In 1906, Alois Alzheimer reported on an institutionalized female patient with progressive dementia. Her first symptoms involved personality changes, but soon after she exhibited a profound memory impairment. After being shown objects and recognizing them, she immediately forgot them and the circumstances in which she had learned about them. Patients with Alzheimer's disease exhibited a set of prototypical symptoms including the cardinal signs of anterograde and retrograde amnesia. Initially, the patients would show mild memory lapses. As the disease progressed, the patients would become profoundly forgetful, remembering things said for only a few minutes and then completely losing them. Consistent with the law of regression Ribot had described for retrograde amnesia following head injury, the impairment in Alzheimer's disease was more severe for memories recently acquired than for those acquired earlier in life. Another disease with prominent loss of memory was first described by Sergei Korsakoff in 1887. His initial report involved a group of patients with an odd combination of peripheral neuromuscular symptoms (polyneuritis) and memory disorder. Many of these patients were chronic alcoholics, who came to the clinic in a global confusional state that gradually resolved, leaving an outstanding selective impairment in memory as the outstanding prominent symptom. The characterization of the memory loss in Klorsakoff syndrome was similar to that for the early stage of Alzheimer's disease. These patients could follow a train of conversation, but, when distracted even for a brief period, they lost both the contents of the conversation and the memory that it had taken place. The patients also showed the signs of retrograde amnesia, including the temporal gradient in which remote memories were more preserved than recent ones. Consolidation and normal human memory

In a monograph published in 1900 Georg Muller and Alfons Pilzecker reported a large number of experiments performed on normal human subjects. They had adapted Ebbinghaus's method for learning "nonsense syllables," short and pronounceable but meaningless character strings, presenting a list of them in pairs and then asking subjects to recall the second item in each pair upon subsequent probing with the first item. A major finding involved the observations of a strong tendency for subjects to

Four Themes in Research on the Neurobiology of Memory

23

spontaneously become aware of the training pairs during the retention phase, even when they tried to suppress rehearsal. They called this phenomenon "perseveration" and linked it to the additional observation that errors made during recall involved items in the same list much more often than those from separate lists. Also, both phenomena of perseveration followed a regular time gradient—they were much more prominent for a few minutes after original learning than later. Muller and Pilzecker speculated that the perseveration reflected a transient brain activity that might play an important role in establishing and strengthening the word associations. They postulated that if this were the case, then the disruption of perseveration should have a deleterious effect on recall. To test their hypothesis Muller and Pilzecker evaluated the effects on recall performance of interpolated material given in between the initial presentations and the recall test. They tested the effects of presenting an additional list in between training and recall on an initial list, finding that indeed recall was poorer if an additional list was presented, as compared to the results with no intervening material. They called this phenomenon "retroactive interference." They varied the nature of the interpolated material by assessing the effects of presenting pictures instead of another verbal list, and found that this distraction was also effective in producing retroactive interference. Furthermore, they varied the timing of presentation of the interpolated material, and found that delaying the distraction by more than a few minutes diminished its interfering effects considerably. These studies led them to the conclusion that there is a brain activity that normally perseverates following new learning and that this activity serves to consolidate the memory. These findings were shortly after linked to the reports of retrograde amnesia in patients with brain insult or damage. Thus, temporally graded amnesia was explained as a disruption of the perseveration process caused by a direct functional interruption of the underlying brain activity. In 1903 William Burham described the effect of brain trauma as disrupting a natural physical process of organization associated with a psychological process of repetition and association, processes that required time to mature. Succeeding decades of progress In the following chapters we explore in greater detail all of the issues raised in this chapter. The plan for the remainder of the book is to follow up on each of the four central themes, one at a time. This might seem to suggest that these issues are entirely independent, but this is very much not the case. The discovered characteristics of conditioned reflexes guided much

24

Introduction

of the succeeding work that unsuccessfully addressed the issue of compartmentalization of memory in the cortex. Conversely, the results of succeeding studies on the nature of cognition in memory also strongly influenced the ultimately successful advances in the compartmentalization of memory functions. And succeeding studies on both the basis of cellular connections and cognitive mechanisms have led to a more sophisticated understanding of processes underlying memory consolidation. So, while I will proceed to separate these themes as a heuristic, the research that guides them, the issues themselves, and the findings on each of them are strongly interrelated. Part I of the book updates you on our understanding of the cellular and molecular bases of memory. Chapter 2 reviews the basic anatomy of physiology of neurons, and shows how these basic principles can be put to use in explaining how memory works in relatively simple nervous systems. Chapter 3 describes parallel successes in understanding the cellular bases of a form of plasticity characteristic of mammalian brain areas, and summarizes research indicating that this form of neural plasticity may be the fundamental mechanism of learning in many more complex brain systems. The Part II of the book builds on the discussion of the nature of cognition in memory. I will update you on how the controversy between behaviorists and cognitivists played out in the middle of the twentieth century, and then how it was resolved by discoveries in neuroscience. In particular I consider a major discovery in the neurology of memory, a case study of amnesia that ultimately showed that memory could be isolated as a cognitive function and that laid the groundwork for resolving the controversy between cognitivist and behaviorist views of memory. Then I elaborate on our understanding of a memory system that mediates "cognitive" or, as it is called today, "declarative" memory. Chapter 4 reviews the evidence from studies of amnesia in humans, and chapter 5 covers the additional evidence from animal models of amnesia. Chapter 6 summarizes complementary evidence from observations on brain activity during declarative memory in humans and animals. In Part III, I summarize progress on the issue of compartmentalization. In Chapter 7, I begin by describing how the controversy over cortical localization became a central issue in research on memory per se, and I show how this controversy was resolved by our modern understanding of cortical modules in information processing and memory. Then I summarize the current psychological, anatomical, and physiological evidence about multiple memory systems in the brain. Chapter 8 introduces substantial direct evidence for the existence and initial localization of multiple mem-

Four Themes in Research on the Neurobiology of Memory

25

ory systems in the brain. Chapter 9 elaborates the anatomy and workings of the full system that mediates declarative memory. Chapters 10 and 11 elaborate on two major systems, one for procedural (habit and skill) learning and the other for emotional memory. In Part IV, I consider progress on the issue of memory consolidation. In Chapter 12, I describe how modern research has distinguished between two different kinds of consolidation, a short-term cellular fixation process and a long-lasting reorganization process. This chapter reviews the evidence for modulation of memory fixation and considers brain mechanisms that mediate memory reorganization. Finally, Chapter 13 returns to an emphasis on a particular part of the cerebral cortex, the prefrontal area, and how this area along with other cortical areas works to orchestrate memory. Throughout the text you will see that the issues laid out a century ago are as relevant today as when they were introduced. But now we are truly beginning to resolve the anatomy and mechanisms of memory at a level of sophistication that could not have been envisioned so long ago. Summing up There are four main themes in studies on the neurobiology of memory: connection, cognition, compartmentalization, and consolidation. Connection concerns the most basic level of analysis of memory function, the fundamental nature of the circuitry of the brain including the elements of information processing and how they communicate with one another in the service of memory. Neurons are independent elements of information processing that are connected via synapses. In the simplest circuits, neurons connect sensory inputs to motor outputs to mediate reflex arcs. However, most reflex circuitries involve more complex arrangements that offer considerable coordination and control over behavior. There are also conditioned reflexes that involve the association of an arbitrary stimulus and an unconditioned stimulus, such that the conditioned stimulus comes to produce a conditioned response that is similar in form to the unconditioned or reflexive response. Conditioning is thought to be mediated by an enhancement or elaboration of the connections between neurons involved in reflex arcs. Cognition refers to the nature of memories at the highest level of analysis, the psychological level. The central issue in the understanding of the psychological nature of memory involves the debate between behaviorism, which espouses that all learning can be reduced to conditioned responses, and cognitivism, which argues that more complex phenomena such as insight and

26

Introduction

inference are required to explain complex learned behavior. Over most of the twentieth century evidence for both views has been accumulated. Compartmentalization refers to the notion that memory as a whole is distributed widely in the brain, and at the same time, there are different kinds of memory that are accomplished by specific brain modules, circuits, pathways, or systems. Gall first proposed the first detailed function mapping of the cerebral cortex, which he called "organology." However, his flawed methods led to incorrect assignments of function-structure relations. Later neurologists discovered case studies of humans with specific cortical damage and consequent specific deficits in language. Also, physiologists demonstrated that specific cortical areas in monkeys had identifiable delimited functional roles in sensory or motor processing. Consolidation is a hypothetical phenomenon derived from the observation that memories are initially labile and later become resistant to loss, suggesting an extended process during which memories take on a permanent form. The existence of consolidation has been shown in studies on patients with damage to the brain showing a temporally graded retrograde loss of memories, that is, intact memories for material acquired remotely prior to the damage and lost memories for materials learned recently prior to the damage. The existence of consolidation can also be observed in normal humans by interposing interfering materials briefly, but not delayed, following initial learning.

READINGS Eichenbaum, H., and Cohen, NJ. 2000. From Conditioning to Conscious Recollection: Multiple Memory Systems in the Brain. New York: Oxford University Press. Finger, S. 1994. Origins of Neuroscience: A History of Explorations into Brain Function. New York: Oxford University Press. Finger, S. 2000. Minds Behind the Brain. New York: Oxford University Press. Lechner, H.A., Squire, L.R., and Byrne, J.H. 1999. 100 years of consolidation— Remembering Muller & Pilzecker. Learn. Mem. 6:77—87. McGaugh, J.L. 2000. Memory—a century of consolidation. Science 287: 248-251. Milner, B., Squire, L.R., and Kandel, E.R. 1998. Cognitive neuroscience and the study of memory. Neuron 20:445-468. Polster, M.R., Nadel, L., and Schacter, D.L. 1991. Cognitive neuroscience analyses of memory. A historical perspective. J. Cog. Neurosci. 3:95-116. Tolman, E.C. 1948. Cognitive maps in rats and men. Psychol. Rev. 55:189-208. Zola-Morgan, S. 1995. Localization of brain function: The legacy of Franz Joseph Gall (1758-1828). Annu. Rev. Neurosci. 18:359-383.

I CONNECTION: THE CELLULAR AND MOLECULAR BASES OF MEMORY

T

he parallel anatomical, physiological, and behavioral studies of Cajal, Sherrington, and Pavlov provided an immensely strong foundation for the conditioned reflex as a central model of the basic memory circuit. And this model became the centerpiece of the biological instantiation of the learning mechanism for behaviorists. In succeeding decades further major advances in our understanding of the basic elements of neural connections would be made. In particular, one major discovery was that the nature of transmission of information between neurons is chemical. The key experiment was performed by Otto Loewi in 1921. He was familiar with current work that had shown that chemical agents could stimulate and modulate the actions of the autonomic nervous system. The general view was that communication across neurons was by an electrical impulse, but the notion that there might be chemical transmission was being considered seriously. Loewi devised an experiment that would provide proof of a chemical mechanism. He removed the heart of a frog and bathed it in a neutral solution, where he stimulated the still attached vagus nerve, producing a well-known inhibition of contractions in the heart. Loewi then removed some of the bathing solution and placed it into a chamber holding another frog heart for which the vagus nerve had been removed. The second heart also slowed, demonstrating that a chemical agent that had been produced in the stimulated heart caused the inhibition. Loewi also performed the complementary experiment, showing that other stimulation that produced an acceleration of the first heart resulted in the release of a chemical agent that also accelerated the second (nonstimulated) heart. It turns out that the inhibitory agent is the neurotransmitter acetylcholine, the major neurotransmitter of the parasympathetic system, and 27

28

Part I: Connection

the accelerating agent is noradrenaline, the major neurotransmitter of the sympathetic system. Since the discovery of neurotransmitters, two major lines of research have refined our understanding of the mechanisms of chemical communication between neurons. First, the pioneering work of Bernard Katz in the 1950s showed that neurotransmitters are released at the synaptic ending of the axon in small packets of molecules called synaptic vesicles. Second, a long list of other neurotransmitters and neuromodulators has now been described, allowing for a range of effects that can be accomplished via neurotransmission. In addition, interest in the molecular and cellular basis of synaptic modification has intersected with the issue of memory consolidation. In the 1960s and 1970s considerable effort was placed on showing that protein synthesis was required for permanent modifications of cells for lasting memory. Many studies showed that interfering with the synthesis of proteins, using drugs that prevent specific stages of gene expression, blocked the establishment of long-term memory without affecting short-term memory. Moreover, these drugs were effective in blocking later expression of memory even if they were given a few minutes after training, but not if treatment was delayed by an hour or more. The results of these experiments paralleled the time course of effects of other types of interference or brain insult that characterized the earlier studies demonstrating memory consolidation. In addition, the studies on protein synthesis inhibition provided a much more specific mechanism, suggesting that gene expression leading to proteins is a critical part of the consolidation process. As you will see in Chapters 2 and 3, this search has now narrowed substantially toward investigations on particular types of neurotransmitters that are activated during learning and on the identification of specific molecular pathways of subsequent gene expression. The following two chapters review the state of our understanding about the cellular mechanisms of memory. In Chapter 2, I summarize our knowledge about the basic anatomy and physiology of neurons. Then I show how these basic anatomical and physiological elements are modified to mediate memory within model systems in invertebrate species. Chapter 3 builds on these observations, introducing a mammalian model system for cellular plasticity that shares many of the features of the invertebrate models and expands on them and other mechanisms within more complex circuitries. Then I consider how well this model works in accounting for cellular mechanisms of real memory in mammals.

2 Neurons and Simple Memory Circuits

STUDY QUESTIONS What are the elements of neuronal structure? How do the electrical properties of neurons arise? What distinguishes different forms of neural conduction and transmission? Why are simple invertebrate systems useful for understanding the cellular mechanisms of memory? What are the molecular and cellular bases of simple forms of learning, including habituation, sensitization, and classical conditioning?

urons encode memories by modifications in the strength of the functional connections. In this chapter I summarize some of the key funN damental concepts about the anatomy and physiology of neurons, include

ing the molecular basis of the unusual electrical properties of neurons, different forms of electrical conduction, and transmission of information between neurons. Then I show how these concepts can be put to work toward understanding the cellular bases of basic forms of learning. Three elemental forms of learning that are accomplished within the circuitry of relatively simple animals have served as model systems for the study of memory. These studies have provided a clear understanding of the mechanisms of neuronal plasticity that mediate habituation, sensitization, and classical conditioning mediated within well-identified circuits of a marine invertebrate. 29

30

Part I: Connection

Neuron structure As recognized at the time of Cajal, neurons are composed of four main elements, the dendrites, the soma or cell body, the axon, and the synapse (Fig. 2-1). There are typically many dendrites and they are often highly branched, such that most neurons receive inputs at the synaptic connections with axons of many other neurons. Each neuron has only one cell body, although the axons of many other cells can contact the cell body directly and these contacts are particularly effective. Each neuron also has only one axon, although in many situations the axon can branch extensively to make a large number of contacts onto one or more other neurons. These basic anatomical facts dictate that neurons receive and integrate information from a substantial number and variety of inputs, and then sum them up to a single main output that can affect one or many cells that are next in the circuit. The synapse is a complicated structure, composed of two main parts, the presynaptic and the postsynaptic elements. The presynaptic element is an enlargement of axonal ending that contains specialized machinery for the process of neuronal transmission. This machinery includes cellular organelles that produce energy and elements that are involved in the recycling and packaging of neurotransmitters into synaptic vesicles. In addition, there are specialized docking stations for the vesicles from which the neurotransmitter is released. There is a narrow separation between the presynaptic element and postsynaptic cell membrane of the neuron to which it is connected; this separation is known as the synaptic cleft. When released from the presynaptic element, neurotransmitters must diffuse

Figure 2-1. Examples of different types of neurons and their major components.

Neurons and Simple Memory Circuits

31

across the synaptic cleft to reach specialized receptors in the postsynaptic element. There are many different kinds of receptors, including distinct types of receptors for the same neurotransmitter, providing for a variety of effects of transmission on the target cell's activity. The structure of these receptor molecules and mechanisms of their activation are only now becoming known. Within the overall design described above, the elements of neurons can take many configurations suited to specific applications in different parts of the nervous system (Fig. 2-1). A prototypical principal neuron is the pyramidal cell of the cortex and hippocampus. These neurons have a long branching dendrite that extends upward from the cell body and receives inputs from other regions, as well as multiple dendrites that branch laterally from the cell to receive inputs from local neurons. The axon extends downward and branches. It may connect with other local cells or extend many millimeters to another brain region. Interneurons are neurons that receive inputs and send their outputs within a local brain region. Motor neurons of the spinal cord have many branching dendrites that extend in all directions, and a single long axon that extends very long distances to innervate skeletal muscles. Its axon branches extensively and has specialized presynaptic terminals to make contact with muscle cells. Sensory cells, conversely, have specialized endings of their dendrites to receive information from specific sensory organs. Some of them may have the cell body displaced such that it is connected to a single main dendritic and axonal element. There are many other variations on these patterns. The physiology of neurons As I describe the physiology of neurons, it should be kept foremost in mind that communication in the nervous system involves three main stages that are mediated by different physiological mechanisms (Fig. 2-2). Two of these stages involve electrical mechanisms for conduction of neuronal signal over substantial distances through the dendrites and axon, and the third involves chemical mediation of transmission of the signal between neurons. The initial phase of conduction, known as electrotonic conduction, typically begins at the postsynaptic site, and proceeds to the cell body. This type of electrical conduction is remarkably fast, but dissipates over relatively short distances. The second type of electrical conduction, called the action potential, is initiated by a special mechanism at the cell body and conducted down the axon to the presynaptic elements. This type of electrical conduction is relatively slow compared to passive conduction, but involves a mechanism that maintains the signal over very long distances. When the action potential

Figure 2-2. Comparison of decremental conduction and action potentials. On the left are idealized waveforms of synaptic potentials recorded at successive loci indicated by the site of the recording electrode. Note that the latency after stimulation until onset of the potential is very short for all recordings. On the right are idealized waveforms of action potentials recorded at successive loci indicated by the sites of recording electrodes. Note that the latency increases substantially between recordings, showing the slow conduction of action potentials as compared with electrotonic conduction.

32

Neurons and Simple Memory Circuits

33

reaches the presynaptic element, the molecular processes of synaptic transmission are initiated and carry a chemical signal across the synaptic cleft to the presynaptic elements of the next cell. In the following sections, I summarize and compare the different forms of electrical conduction and then describe the mechanisms of synaptic transmission. The resting potential To understand the mechanisms of electrical conduction it is important to appreciate that nerve cells, as well as other types of cells throughout the body, have a natural electrical potential known as the resting potential. This potential arises from two features common to most living cells (Fig. 2-3). First, there is a natural concentration of molecules inside the cell

Figure 2-3. Schematic diagram of ions and their flow through the cell membrane (dashed lines) associated with the resting potential, synaptic potential, and action potential.

34

Part I: Connection

membrane relative to the fluid outside the cell (called the extracellular fluid), and many of these molecules are polarized in that they have a positive or negative charge. Major contributors to the high intracellular (i.e., inside) concentration are large protein molecules that have a negative charge that is balanced out by small positively charged molecules such as sodium (Na + ) and potassium ( K + ) in the intracellular fluid. Second, the membrane that surrounds the cell and contains its contents is typically somewhat permeable in that it contains pores or channels that allow the diffusion of molecules between the inside and outside of the membrane. These two situations conspire to create a situation in which a potential difference between the inside and outside of the cell membrane is established. The potential difference across the membrane arises because of two natural and competing forces that are consequences of these properties of neurons. One of these forces comes about because of the difference in concentration of molecules inside and outside the membrane. The accumulation of large molecules inside the cell creates a substantial concentration gradient, such that the interior fluid of the cell is much more concentrated than the fluid of the exterior of the cell. If the membrane pores or channels were indifferently permeable to all molecules, the difference in concentration would disappear as molecules from the inside diffused out until the concentration on both sides of the membrane was equal. However, the membrane channels are actually quite selective, typically allowing small molecules such as Na+ and K+ to flow, and even the passage of those molecules is tightly regulated. Indeed, in the natural resting state, the cell membrane is mostly permeable only to K+. Because of this selective permeability and because there is typically much more K+ inside the cell than out, some K+ flows from inside to outside following the concentration gradient, and this creates a situation where the inside of the cell has a net negative charge due to the loss of some positively charged potassium molecules. If the concentration gradient was the only force involved, K+ would continue to flow outside the cell until its concentration was equal on both sides. This overall potential difference, with the inside negatively charged and the outside positively charged, invokes the second force that affects the resting potential. The interior-to-exterior charge difference constitutes an overall electrostatic gradient across the cell membrane that tends to repel positively charged molecules from the positively charged exterior. At the same time, because the interior of the cell has become negatively charged, there is an overall attraction of positive charges to the inside of the cell. Because the cell membrane is selectively permeable to K+, and so it is the

Neurons and Simple Memory Circuits

35

only molecule that can pass, the combination of repellant force from the outside and attractive force from the inside draws some K+ molecules back into the cell. If the electrostatic gradient was the only force involved, K+ would flow back into the cell until there was no charge difference. Thus, there are two competing forces on the K+ molecules. The concentration gradient pushes K+ outside, whereas the electrostatic gradient pulls K+ inside. These two forces eventually reach an equilibrium in which K+ molecules leave the cell to decrease the concentration gradient only somewhat, resulting in an overall net electrostatic gradient in one direction and a concentration gradient of equal force in the opposite direction. This final electrostatic gradient is called the resting potential, and its magnitude is typically about —70 millivolts (inside relative to outside the membrane). The exact magnitude of this potential depends on many factors, including the concentrations of K+ and other molecules for which the cell membrane is typically less permeable. Electrotonic conduction In the natural situation the resting potential is disturbed when chemical processes at the synapse result in a change in the permeability to a molecule, typically Na+ (Fig. 2-3). Unlike K+, Na+ is more concentrated outside the cell membrane. This is because there is an active mechanism, embedded in the cell membrane, that pumps Na+ molecules from the inside to the outside of the cell. This pump is metabolically expensive, consuming as much as 40% of the energy needs of a neuron. But it is a very valuable mechanism for the conduction of signals. When synaptic transmission results in a transient increase in permeability to Na+ at the receptor site, Na+ follows its concentration gradient and flows into the cell. Because Na+ is positively charged, this results in movement of the membrane potential in the positive direction, from —70 millivolts to something closer to zero, with the magnitude depending on the strength of the synaptic transmission. This relative positive charge gradient spreads passively and almost instantaneously, like electricity, decreasing the polarization of the membrane for some distance along the dendrite toward the cell body. However, as this depolarization spreads, it also diffuses across the membrane surface and so diminishes in size. Thus, electrotonic conduction is fast but it is also decremental, in that it decreases in magnitude so that a smaller depolarization reaches the cell body (see Fig. 2-2). The size of the potential depends critically on the distance it must travel, such that synapses on very distant dendrite branches are usually much less effective than those near the cell body.

36

Part I: Connection

The action potential Something truly magical happens at one end of the cell body, and it involves a mechanism that defines the special physiology of the neuron. This special property, which involves a dramatic alteration in the membrane permeabilities, typically occurs only in axons. At the postsynaptic site in dendrites and in most of the cell body, the membrane does not generally change its permeabilities substantially. So, potentials created at the synapse are propagated only by electrotonic conduction of the transient perturbation of the resting potential at the synapse when transmission occurs. As described previously, this perturbation is conducted rapidly, at the speed of electricity, but it also decrements rapidly along the length of the dendrite. So electrotonic conduction is not a mechanism that would support long-range communication of signals down the axon for distances of up to a meter or more, as required in some pathways of the brain. At the end of the cell body where the axon originates, the membrane takes on a profound new quality, one that allows it to change its permeabilities in a large and very useful way. At this locus, called the initiation zone, the channels in the membrane change their permeability when they are depolarized to a specific threshold, and hence they are called voltage-gated channels. The mechanisms of these channels in mediating the action potential were first discovered by Alan Hodgkin and Andrew Huxley in the 1940s. They showed that depolarization of the axon membrane above the threshold, typically about a 15-20 millivolt depolarization, results initially in an increase in permeability selectively to Na+ molecules. As was the case at the synapse, Na+ is in greater concentration outside the axon, and therefore an increase in Na+ permeability results in an influx of positively charged sodium molecules, and consequently a further depolarization of the cell membrane. This sets up an unusual "regenerative" situation, a positive feedback mechanism by which the initial depolarization above threshold causes an influx of Na + , which further depolarizes the cell, which causes more Na+ channels to open, which allows more Na+ in, which additionally depolarizes the cell, and so on. When does this regenerative loop end? Sooner or later, the membrane becomes fully permeable to Na+ and it will reach its equilibrium potential, that is, its balance between concentration and electrostatic gradients. Thus, the maximum potential is a fixed number for a given cell (whose value depends on factors such as the initial inside and outside Na+ concentrations, and the temperature). Because the pump makes the Na+ concentration greater outside the cell, its equi-

Neurons and Simple Memory Circuits

37

librium is reached when the inside of the membrane reaches an overall positive value, typically about 40 millivolts. Importantly, once the threshold of depolarization is achieved, this regenerative process runs itself inevitably and precisely to the equilibrium potential for Na + . The magnitude of the action potential is much larger than the potentials associated with decremental conduction, and its appearance is "all-or-none." Either the threshold for activation of the voltage-gated channels is not reached, and very little potential is obtained and dissipates rapidly, or threshold is reached, and the mechanism regenerates itself up to the full value of the equilibrium potential for Na + . Furthermore, when an action potential is generated in the initiation zone, the potential spreads electrotonically. This spread would decrement, but over a short distance would be more than sufficient to take the adjacent voltage-gated channels of axon membrane to threshold. This would regenerate the full magnitude of the action potential at that neighboring locus, and that potential would itself spread, reinitiating a full-blown action potential at its neighboring loci, and so on, continuing to reduplicate the action potential through the length of the axon. This simple regenerative mechanism, therefore, not only insures that the action potential achieves its full size at each locus but also insures its propagation for the full length of the axon regardless of the distance involved. The action potential also includes a mechanism for recovery. Shortly after the Na+ channels are activated, voltage-gated channels for K+ also open. During the resting phase, K+ channels were open to some extent, allowing the establishment of the resting potential. In addition, just after the initiation of the Na+ current, other K+ channels are activated, allowing even greater permeability to this molecule. Because K+ is more concentrated on the inside of the cell, it flows out, and does so especially strongly because the Na+ onrush has made the inside of the cell move to a positive potential. The result of this series of events is that the rise in the membrane potential to a positive state is short-lived. The membrane potential rapidly returns to its initial level (indeed to even below that level—an overshoot—because of the especially high K+ permeability). At that point the membrane has more or less reachieved its normal potential but the molecular balance is not the same as its initial status. There is a lingering high concentration of Na+ inside the cell, and extra K+ has left the cell. The pumping mechanism sets this imbalance right, but during this recovery period a new action potential cannot be initiated, and so the cell is said to be in a refractory period.

38

Part I: Connection Synaptic transmission

When the action potential reaches the end of the axon, at the presynaptic site, another mechanism takes over to mediate synaptic transmission (Fig. 2-4). The spreading of the action potential to the membrane of the presynaptic element activates voltage-gated channels in that area for a different molecule, calcium (Ca 2+ ). Calcium is more highly concentrated on the outside of the cell, in the synaptic cleft, and so the depolarization of the presynaptic element results in a substantial influx of Ca2+ into the presynaptic element. It takes time for these channels to open, in part accounting for the delay in synaptic transmission of signals, but Ca2+ is the critical catalyst to initiate synaptic transmission. It appears that Ca2+ plays a central role in facilitating the docking of synaptic vesicles at specific sites in the end of the presynaptic membrane. When this docking has been accomplished, the vesicle fuses with the end of the cell membrane and re-

Figure 2-4. Schematic diagram of events in synaptic transmission.

Neurons and Simple Memory Circuits

39

leases its contents, the neurotransmitter, into thhe synaptic cleft. Quickly the vesicular material is recaptured and recycled, to be filled again with neurotransmitter. Meanwhile, the neurotransmitter diffuses across the synaptic cleft and finds its way onto and binds with specific receptors. The receptors briefly bind the neurotransmitter, and in doing so, activate channels that control the permeability of the postsynaptic membrane to local molecules. The binding is typically transient, interrupted by an unbinding and diffusion of the neurotransmitter, or by other molecules that destroy the neurotransmitter. The neurotransmitter or its breakdown products are not allowed to remain in the synaptic cleft for long, however. They are reabsorbed by the presynaptic membrane and recycled. One main action of these receptors is to initiate changes in the postsynaptic membrane already described, the opening of Na+ channels to initiate an excitatory postsynaptic potential that can be propagated electronically along the dendrite. However, there are many types of neurotransmitters and many types of receptors, even for the same neurotransmitter molecule. This provides considerable capacity for regulation of the duration and type of potential produced at the postsynaptic element. One major variant is the capacity to generate hyperpolarization, rather than depolarization, of the postsynaptic element. This is accomplished, at least in some of these inhibitory synapses, by transmitter-gated chloride channels in the postsynaptic element. Chloride is a negatively charged molecule that is more concentrated outside the cell membrane. When its channels are activated, the negatively charged chloride molecules flow inside, making the postsynaptic element become even more polarized than normal. This hyperpolarization also flows via electrotonic conduction across the dendrite and can serve to inhibit the generation of an action potential in the activation zone. Integration of synaptic potentials In most cases the synaptic potentials that are initiated by synaptic transmission are relatively small, about 5-20 millivolts. As described earlier, these subthreshold potentials decrement in magnitude as they travel to the initiation zone in the axon. Therefore, in most cases, an action potential does not occur as a result from a single activation of one synapse. Instead, the initiation of an action potential typically requires summation of many synaptic potentials. This summation can occur across time if the same synapse is activated repeatedly quickly enough so that the synaptic potentials can build up. Also, summation can occur by the concurrent acti-

40

Part I: Connection

vation of many spatial distant synapses on the same dendrite or different dendrites. The combination of excitatory (depolarizing) and inhibitory (hyperpolarizing) synaptic potentials offers considerable fine tuning of the likelihood of an action potential. In addition to these forms of temporal and spatial summation, the degree to which individual synapses control the action potential depends to a great extent where on the dendrite it is initiated—synapses close to or on the cell body are most effective, because they will suffer less from the effects of decremental conduction, as described earlier. These aspects of integration are complemented by a variety of mechanisms that regulate the efficacy of synapses and their consequent contribution to initiation of action potentials. The efficacy of a synapse is determined by the supply of neurotransmitter, and by the amount and duration of Ca2+ influx that determines how much neurotransmitter is released. The sensitivity and duration of receptor activation can be regulated by substances in the synaptic cleft, and, indeed, many drugs operate by interfering with or enhancing the operation of receptors. The number and sensitivity of receptors also determine the efficacy of the synapse. As you will see, alterations in each of these parameters provides a mechanism for changes in synaptic efficacy that underlie memory storage. Cellular biology of simple memory circuits To provide examples of how the cellular and molecular mechanisms of neural conduction and transmission become important in memory, the remainder of this chapter summarizes a program of study on the behavior and physiology of a relatively simple invertebrate species. Invertebrates are superb animals in which to study the cell biology of nervous function because their nervous systems involve many fewer neurons than those of most vertebrates, and many of their neurons are quite large and unique. These qualities allow researchers to individually identify exactly the same set of cells in each animal, and to study virtually all of the major cells involved in a functional circuit. In pioneering studies, Eric Kandel and his colleagues have examined the behavioral, anatomical, and physiological properties of simple forms of memory in Aplysia, a large sea snail. Most of their studies have focused on: one particular reflex circuit, called the gill withdrawal reflex. When the snail is quiescent, it extends its gills from the abdominal region, as well as a fleshy continuation of the gill called the siphon. Ordinarily the siphon serves to assist the flow of aerated water over the gills as they function in respiration. However, the gills are a delicate organ, one that is easily dam-

Neurons and Simple Memory Circuits

41

aged. Therefore, when there are signs of danger, the snail can withdraw the gill and siphon. In the laboratory, a mantle shelf that ordinarily completely covers the gill and siphon can be retracted, and then when the animal is relaxed, the gill and siphon are extended. The defensive reflex is initiated by a gentle stroke of the siphon with a paint brush. This results in the rapid withdrawal of the gill under the mantle shelf. This reflex has been studied extensively, using behavioral paradigms that examine three simple forms of learning, habituation, sensitization, and classical conditioning. For each form of learning the anatomical circuit of the relevant neurons has been characterized, and the cellular and molecular mechanisms that underlie learning have been explored. A review of the findings of these studies provides a solid introduction to central principles of the cellular mechanisms that mediate learning. Habituation Perhaps the simplest form of learning is habituation. All of us use habituation every day to help us learn not to attend or respond to irrelevant stimuli. For example, if you ever moved from a small to a big city, your attention to the noise of traffic may have initially made it difficult for you to sleep through the night. Each time a car blew its horn or a siren went off, you woke up. However, after several days, you probably came to ignore the noises, that is you habituated to them, and your sleep was undisturbed by them. Rapid and lasting habituation can also be observed in the gill withdrawal reflex of Aplysia. Following elicitation of the reflex and subsequent relaxation, if the siphon is stimulated again the reflex is smaller, and following several repetitions it becomes quite reduced in magnitude. Furthermore, following only 10 stimulations, the reflex may remain habituated for only 15 minutes. But, after 4 days of such training, the habituation can last weeks. The longer-lasting habituation is a very simple form of learning, to be sure, but it has the lasting property that indicates it is indeed a form of long-term memory. Now the researchers sought to characterize the circuit of nerve cells involved in the reflex and to determine the mechanism of lasting habituation. A highly schematic sketch of this circuit that shows just one of each type of cell is provided in Fig. 2-5. It turns out that all the relevant cells of the circuit are in a single ganglion, or cluster of cells in the abdomen. There are about 40 sensory cells that innervate the siphon skin, and these connect with six motor neurons that innervate the gill musculature. The neurotransmitter for this synapse is glutamate, and this will become important later. There are also clusters of both excitatory and inhibitory in-

42

Part I: Connection

Figure 2-5. Schematic diagram of an idealized circuit for habituation in Aplysia, and recordings of action potentials at successive stages in the circuit.

terneurons, cells that receive input from the sensory cells and innervate the motor neurons. When the stimulus is initially applied to the siphon, the sensory neurons are excited, and these excite the interneurons and gill motor neurons. These inputs converge and cause the motor neuron to discharge action potentials repeatedly, producing a vigorous withdrawal response. Following repetitions of the siphon stimulation, the sensory neurons still produce action potentials, indicating that the habituation is not mediated by a sensory adaptation or any other change in the responsiveness of the sensory elements (Fig. 2-5). However, the magnitude of the synaptic potential in the interneurons and motor neurons is reduced, such that the likelihood of generating an action potential in the motor neuron, and the number of them generated, is smaller. Eventually, even though the sensory neuron is still responding vigorously, no action potentials are produced in the motor neurons. This means that the locus of the memory is to be found in the physiology of the connections between the sensory neurons (as well as the excitatory interneurons) onto motor neurons. This was confirmed by closely examining the synaptic potentials in the motor neurons and confirming that the magnitude and longevity of their depression closely mirror that of the behavioral response. Now, Kandel and colleagues asked what stage in the process of conduction or transmission was affected by habituation. They found that the

Neurons and Simple Memory Circuits

43

receptors in the postsynaptic site were equally sensitive, so the locus of the depression had to be presynaptic. Indeed, they found that the depression was due to a decrease in the number of synaptic vesicles released for each action potential. They then used an electron microscope to examine the number and locations of synaptic vesicles in the presynaptic element before and after habituation, and found that the number of available vesicles did not decrease but fewer of them became docked onto release sites in habituated animals. Furthermore, following extended training, physiological assessments indicated that many fewer sensory neurons had effective connections with motor neurons, and anatomical examination showed that the number of synaptic contacts between sensory neurons and interneurons and motor neurons was substantially reduced. These findings showed that memory can be mediated by changes in synaptic efficacy, both through intracellular mechanisms that control transmitter release and through changes in anatomical connectivity. Sensitization A second simple form of nonassociative learning observed in Aplysia is sensitization. This kind of learning is, in a way, the opposite of habituation in that it involves an increase in reflex magnitude as a result of stimulation. In this case though, the circuit involves a combination of two inputs such that strong stimulation of one input sensitizes, or makes more vigorous, responses to the other input. An everyday example is when we encounter a fearful stimulus, such as a loud noise, we become for some time more likely to startle, or startle more vigorously, to many other sounds as well. In Aplysia, sensitization has been studied using a protocol in which initially the tail of the animal is stimulated with an electric shock, which results in an increase in the robustness of the gill withdrawal to siphon stimulation. A single tail shock produces sensitization that lasts for minutes, whereas a series of 4-5 tail shocks produces sensitization that lasts for a few days. The circuit that mediates this form of learning involves the same set of cells involved in habituation, plus sensory neurons that innervate the tail and additional interneurons (Fig. 2-6). Thus, one way or another, the same set of cells can mediate both habituation and sensitization, two different forms of learning. In the case of habituation, the synaptic depression that occurs is said to be homosynaptic, because the mediating events occur within the same pathway that constitutes the reflex itself. However, in the case of sensitization, there must be a facilitation that is mediated by a heterosynaptic mechanism because it involves modulation of the reflex pathway by another set of cells, in this situation in the tail sensory pathway.

44

Part I: Connection

Figure 2—6. Schematic diagram of an idealized circuit for sensitization in Aplysia, and recordings of action potentials at successive stages in the circuit.

The central events that underlie sensitization do not involve direct modifications of the siphon sensory neuron, as was the case in habituation. Rather, the tail shocks activate sensory neurons for the tail, which in turn activate modulatory interneurons that synapse onto the cells bodies and presynaptic elements of the siphon sensory neurons. These modulatory neurons affect the strength of synaptic signal produced by the siphon sensory neurons, and so use an indirect mechanism to support the facilitation of reflexes. Specificially, the facilitation is due to forms of modulation that act to increase the number of synaptic vesicles released by the siphon sensory neuron onto its targets, resulting in a substantial increase in the response of the gill motor neurons. The key to the modulation of sensory neuron synaptic potentials lies in a distinction between two types of receptors (Fig. 2-7). As described before, the conventional receptors, called ionotropic receptors, are found in postsynaptic elements, are transmitter-gated, and allow charged molecules to flow briefly inducing the postsynaptic excitatory and inhibitory potentials. There is also a second class of receptors, called metabotropic receptors. These are activated by transmitters or other molecules, but do not open channels and directly cause changes in the membrane potential. Rather, they produce other changes in the cell that can have lasting effects on its responsiveness. The changes resulting from metabotropic receptor activation typically involve a cascade of molecular events. Thus, when a neurotransmitter binds

Neurons and Simple Memory Circuits

45

Figure 2-7. Schematic diagram of cellular events that mediate long-term changes in synaptic efficacy of the presynaptic site in Aplysia.

onto a metabotropic receptor, an enzyme is activated which in turn alters the concentration of an intracellular signaling molecule called a second messenger (distinguishing it from the neurotransmitter as the "first" messenger). In this and many other situations the second messenger is cyclic adenosine monophosphate (cAMP), which is synthesized by the enzyme from the common metabolic molecule adenosine triphosphate (ATP). The action of cAMP is to turn on a number of cellular processes through activation of a special protein called cAMP protein kinase, also known as PKA. In turn, PKA mediates its effects by adding a phosphate group to a variety of proteins, activating them to play any of a variety of roles in cell regulation. Thus, the second messenger signaling system is different from the primary synaptic mechanism in having a variety of long-lasting effects. In the case of sensitization in Aplysia, the specific neurotransmitter of the modulatory interneurons is serotonin, which acts on the metabotropic receptors of the sensory neurons to increase their intracellular cAMP,

46

Part I: Connection

which in turn activates PKA. This was found by showing that direct application of serotonin is sufficient to activate cAMP, and direct intracellular injection of cAMP is sufficient to enhance transmitter release by the sensory neuron, and to induce the facilitation of the reflex. Furthermore, intracellular infusion of the main subunit of PKA also produces the facilitation, and inhibiting the action of PKA blocks the facilitation. Thus, the full cascade of events associated with the second messenger is both sufficient and necessary for mediation of sensitization. Furthermore, a key specific effect on the physiology of the synapse has been identified. Both cAMP and PKA (or some of its subunits) act to close one of the K+ channels in the presynaptic membrane of the sensory neuron, reducing the action of this channel in terminating the action potential. This results in a broadening of the action potential, opening the Ca2+ channels of the presynaptic element for a longer period, and allowing a greater amount of Ca2+ to enter the presynaptic element. The increased Ca2+ concentration in the presynaptic element increases the number of vesicles docked per action potential, increasing the release of neurotransmitter, which, of course, leads to a greater response of the motor neurons. Classical conditioning

Habituation and sensitization are considered very elementary forms of learning, because they do not involve the acquisition of an association between stimuli, but rather a change in responsiveness to repeated stimulation of one kind. The simplest form of associative learning is classical conditioning, the kind of learning that was the focus of the pioneering studies of Pavlov described in Chapter 1. In this form of learning two different stimuli are presented in close temporal proximity, such that typically a form of stimulation that does not ordinarily produce a response is presented before another stimulus that does produce the response. After multiple pairings the first stimulus acquires the ability to produce the response. In that sense, classical conditioning involves the acquisition of an association between the first, or conditioned stimulus, and the second, unconditioned stimulus. In Pavlov's dogs the conditioned stimulus was a tone, that did not initially elicit salivation. It was presented repeatedly prior to injection of food into the mouth (the unconditioned stimulus), which did directly elicit salivation. After several pairings, the tone came to elicit the conditioned response of salivation. In the Aplysia model, a protocol for classical conditioning was established using an elaboration of the habituation and sensitization paradigms and their neural circuits (Fig. 2-8). Added to the already described ele-

Neurons and Simple Memory Circuits

47

Figure 2-8. Schematic diagram of an idealized circuit for classical conditioning in Aplysia, and recordings of action potentials at successive stages in the circuit.

ments is an additional reflex pathway from the sensory neurons of the skin on the mantle shelf that connect to the same set of motor neurons that withdraw the gill. This pathway is essentially the same as the pathway from the siphon to the gill, and offers the ability to differentially condition one of those paths, and not the other, consistent with the selectivity of associations in classical conditioning in mammals. In this differential classical conditioning protocol, then, on some trials the mantle shelf is lightly stimulated and then the tail is shocked. The mantle shelf stimulation is referred to as the positive conditioning stimulus (CS+) and the tail stimulation is the unconditioned stimulus (US). On other trials, the siphon is stimulated and no tail shock is given, and this stimulation is referred to as the CS—. The animals come to have vigorously enhanced withdrawal responses to the mantle stimulation, the conditioned response (CR), but not to the siphon stimulation, that is, they become differentially conditioned to the mantle conditioned stimulus. As is the case in mammalian examples of classical conditioning, timing is everything. In this case strong tail stimulation simply produces a generalized sensitization. But in the case of classical conditioning, the protocol produces differential conditioning for the paired stimulus. How is

48

Part I: Connection

this accomplished within the cells of this circuit? The critical steps occur both in the presynaptic and postsynaptic elements of the reflex circuit. At the presynaptic element, it turns out that when an action potential reaches the site, the influx of Ca2+ not only induces transmitter docking and release, but Ca2+ also binds to the protein calmodulin. The calcium-calmodulin complex binds to the enzyme (adenyl cyclase) that generates cAMP. Additionally, bound to calmodulin, the adenyl cyclase becomes more sensitive to activation by serotonin, consequently releasing a greater amount of cAMP and subsequently enhancing the influx of Ca2+. Thus, the critical timing involves first a priming step in which the CS+ (mantle shelf stimulation) generates calcium-calmodulin bound to adenyl cylase, and then the closely following US (tail stimulation) acts via the modulatory interneurons to induce cAMP and PKA responses—the closely timed combination produces especially large cAMP responses only in the conditioned stimulus sensory neurons. In addition, there is a change in the postsynaptic element that also mediates the conditioned response. As you should recall, the neurotransmitter for the reflex pathway is glutamate. This neurotransmitter activates two types of ionotropic receptors on the postsynaptic elements. One is a conventional receptor that regulates Na+ influx. The other is a special receptor, called the N-methyl-D-aspartate (NMDA) receptor, that regulates Ca2+ flow. Under normal operation of the reflex, and under habituation and sensitization, the NMDA receptor is blocked by another charged molecule, magnesium (Mg 2+ ). However, NMDA receptors have the unusual property of being modulated by the voltage of the cell membrane such that when the membrane is depolarized the magnesium block is eliminated and Ca2+ can flow into the cell. During the protocol for classical conditioning this is accomplished when the CS+ activates the conventional ionotropic receptors, allowing Na+ influx and producing the typical synaptic depolarization. This transiently unblocks the NMDA channels, so that if a US occurs briefly afterward, facilitating the synapse so that the motor neuron generates a long train of action potentials, the NMDA receptor opens allowing in Ca2+. This results in a cascade of molecular events that results in lasting modifications of the postsynaptic element. The NMDA receptor and its role in memory in mammalian systems is considered again in greater detail in the next chapter. Substrates of permanent memory traces in cellular mechanisms The sensitization and classical conditioning studies so far described have focused on the local mechanisms within synaptic elements that mediate

Neurons and Simple Memory Circuits

49

short-term memory. To create long-lasting memories, it has long been recognized that additional mechanisms are likely required—mechanisms that rely on structural changes involving the growth of existing synapses or the addition of new synapses. These changes surely involve protein synthesis. Early experiments showed repeatedly that inhibition of protein synthesis, at any stage in the activation of the genes to the production of protein molecules, blocks the formation of memories. Furthermore, the role of protein synthesis in memory formation is a somewhat prolonged one, as demonstrated in studies showing that inhibition of protein synthesis for several minutes after learning also blocks later expression of memory. Following that period, however, inhibition of protein synthesis does not block memory, showing that the process is completed in due course. What are the mechanisms that connect the activation of intracellular mechanisms at the synapse to the production of proteins and the fixation of memory? There is evidence that the first steps begin with the molecular cascade already described (see Fig. 2-7). For example, in the sensitization protocol described earlier, single stimulations with serotonin result in the release of a small amount of cAMP, sufficient to generate small amounts of PKA that have local effects. But the amount of PKA is not sufficient to diffuse in quantity to the nucleus of the cell where the genetic machinery lies. However, repeated stimulations raise the concentration of cAMP to a level where it interacts with the PKA to break up its form into separate functional units. One type of unit, called the catalytic subunit, then diffuses in sufficient quantity to the nucleus to activate the genetic decoding machinery. When the active unit of PKA translocates to the nucleus it phosphorylates (adds a phosphorous-containing molecule) to several factors that activate the transcription of RNA from the DNA code (see Fig. 2-7). The most relevant of these for our purposes is a protein called cAMP-response element binding protein (CREB). One form of CREB, called CREB-1, binds to a special element on DNA called CRE and switches on the genes that code for molecules critical to the fixation of memory. This was confirmed in experiments showing that inhibiting the binding of CREB-1 prevents sensitization whereas infusing phosphorylated CREB-1 into the sensory neuron produces lasting facilitation. This combination of findings indicates that CREB-1 is both necessary and sufficient for the fixation of this simple form of memory. In addition, there are mechanisms for fine tuning the process of fixation. In particular, within the same sensitization paradigm, another form of CREB, called CREB-2, has been identified. CREB-2 appears to suppress CREB-1 and therefore acts as an inhibitory transcription regulator, that

50

Part I: Connection

is, a repressor. Evidence indicates that CREB-2 is regulated by a different protein kinase called mitogen-activated protein kinase, or MAP kinase. MAP kinase acts by preventing CREB-2's blockade of CREB-1, and thus activation of MAP kinase facilitates CREB-1 and memory. This was shown in an impressive demonstration of enhanced learning in Aplysia. Whereas a single stimulation of the sensory neuron ordinarily produces only a shortlasting sensitization, following a treatment that blocked CREB-2, the facilitation was lasting. The precise set of genes activated by the CREB mechanism, and their specific actions in mediating memory fixation, are not well understood. Because of the rapidity of the gene expression following stimulation, it has been suggested that the critical steps may involve a special class of genes called immediate-early genes that are activated quickly and transiently. Within the Aplysia sensitization model, there is evidence that one of these genes may control the production of an enzyme called ubiquitin hydrolase, which appears to diffuse back toward the synapse where it helps release the active subunits of PKA and consequently produce a lasting alteration in synaptic physiology. In addition, there must be a production of as yet unknown molecules that regulate growth of synapses. It has been shown, for example, that long-term sensitization results in a major increase in the number of synaptic terminals of the sensory neurons. Dendritic processes also grow to accommodate the increase in synaptic contacts. Thus, there has to be a coordination of presynaptic and postsynaptic growth that would mediate long-lasting changes in efficacy of the reflex. Other "simple" systems Parallel observations on other model systems indicate that the cellular events and molecular cascade described for Aplysia are conserved in evolution. In particular, one prominent model based on another invertebrate often used in genetic studies provides substantial converging evidence of the scheme outlined above. These studies involve a kind of olfactory learning in fruit flies. In this behavioral paradigm, groups of animals are placed in chambers that contain a particular odor and then shocked briefly. On other trials, the same animals are placed in a different chamber that contains another odor and no shock is given. Finally, the animals are placed in an apparatus that allows the animal to migrate between the two familiar chambers and they typically express memory as a preference for the safe chamber. Several different mutant flies have been tested in this protocol, and a number of them turn out to have specific amnesic deficits.

Neurons and Simple Memory Circuits

51

One of these flies, called dunce, had a defect in the gene that regulates cAMP and had a severe learning deficit. Another with a defect in adenyl cyclase, the enzyme that synthesizes cAMP from ATP, is called rutabaga (one of several named after vegetables, as a reflection on their intelligence), and other flies with defects in genetic regulation of PKA also show deficient memory capacity. Recently, in this model, the role of CREB has also been examined. These studies found that a genetic manipulation that led to overexpression of a CREB repressor blocked the fixation of memory. More strikingly, and similar to the studies on Aplysia, overexpression of a CREB activator reduced the training required to have lasting memory. Usually a single training trial with each odor is sufficient to produce only a shortlived memory. However, when the CREB activator was overexpressed, a single training trial was sufficient to fixate the memory. Summing up Neurons are composed of three main elements: dendrites that are specialized for receiving signals from other cells, the cell body, and the axon that is specialized for conduction of the neural impulse. In addition, there are specialized areas of these cellular components that mediate communication between cells, called synapses, each composed of a presynaptic element where neurotransmitters are stored and released and a postsynaptic element where there are receptors that recognize the neurotransmitter and generate signals in the postsynaptic cell. There are two electrical mechanisms for conduction of the neuronal signal over substantial distances through the dendrites and axon. The initial phase, called electrotonic conduction, typically begins at the postsynaptic site and proceeds to the cell body, and involves passive and fast, but decremental conduction of an electrical signal. The later phase is called the action potential, which typically begins at the origin of the axon, and involves active and relatively slow, but faithful conduction of a signal over long distances. Neural transmission occurs at the synapse and involves the action potential causing the fusion of synaptic vesicles and release of neurotransmitter from the presynaptic element. The neurotransmitter activates voltage-gated channels in the postsynaptic receptor site, which depolarizes the postsynaptic cell in excitatory synapses and hyperpolarizes the postsynaptic cell at inhibitory synapses. These basic mechanisms of neuronal physiology are important to understanding the nature of neural plasticity that underlies memory. These and other aspects of the molecular physiology of neurons have been put

52

Part I: Connection

to use in simple invertebrate systems to understand three fundamental types of learning: habituation, the decrementing of responsiveness to repeated sensory stimulation without reinforcement, sensitization, the incrementing of responsiveness to sensory stimulation following strong stimulation, and classical conditioning, the association of an arbitrary external stimulus with a stimulus that produces a reflexive response. The findings from the studies of relatively simple invertebrate learning models have provided fundamental insights about the representation of memories in the brain of all animals. These basic insights can be summed up as follows: First, Cajal was right in his conjecture that alterations in synaptic efficacy provide the basic cellular mechanism for memory. Second, the nature of the chemical mechanisms involved in memory are not unique "memory molecules," but rather involve a set of adaptations of natural molecular mechanisms by which synaptic activity is regulated. Indeed, the most common mechanisms involve clever uses of ubiquitous molecules, such as cAMP and the genetic code regulators. Third, the changes that mediate memory do not involve special "memory cells," but rather the same cells that perform sensory, modulatory, and motor functions in the reflex pathway. Fourth, and closely related to the last point, the cells of the nervous system seem to be highly adaptable. It takes very few activations of the right type to induce an adaptation of the cellular mechanisms to support memory, and not many more to incite the permanent fixation process. Fifth, and finally, several mechanisms are employed in memory formation at the cellular level, allowing for different forms of memory to be encoded within the same cells, and allowing for a variety of ways to fine tune the memory and its time course at the various stages in a cascade of cellular events.

READINGS Bear, M.F., Connors, B.W., and Paradiso, M.A. 2000. Neuroscience: Exploring the Brain. New York: Williams & Wilkins. Carew, TJ. 1996. Molecular enhancement of memory formation. Neuron 16:5-8. Squire, L.R., and Kandel, E.R. 1999. Memory: From Mind to Molecules. New York: Scientific American Library. Tully, T., Bowling, G., Chistensen, J., Connoly, J., Delvechhio, M., DeZazzo, J., Dubnau, J., Jones, G., Pinto, S., and Regulski, M., et al. 1996. A return to the genetic dissection of memory in Drosophila. Cold Spring Harbor Symposium in Quantitative Biology 61:207-218.

3 Cellular Mechanisms of Memory: Complex Circuits

STUDY QUESTIONS What is LTP? Why is LTP a good model for the plasticity that underlies memory? What are the cellular mechanisms for the induction of LTP? What are the molecular mechanisms for the preservation and expression of LTP? What is the evidence that something like LTP occurs during learning? What is the evidence that the mechanisms of LTP are required for learning?

T

he previous chapter showed how understanding cellular and molecular mechanisms can provide clear insights into the bases for memory in relatively simple nervous systems. Indeed, to the extent that the most important aspects of the relevant circuitry have been included in those model systems, it might not be too vain to conclude that we truly understand how memory works in those circuits. In the present chapter, we aim higher: Can we also understand the nature of learning and memory in more complex systems, such as those of mammals, through a characterization of cellular and molecular properties in the key brain areas involved in memory functions? There is certainly a long history of the expectation that a simple reflex modification mechanism will be conserved across species and in complex as well as simple systems. Pavlov and Sherrington, from distinct behavioral and physiolog53

54

Part I: Connection

ical perspectives, recognized that modification of synaptic function— synaptic plasticity—is the basic substrate of the conditioned reflex mechanism in mammals. In 1949 Donald Hebb broadened this notion beyond that of stimulus and response associations, outlining a cascade of events that begins with synaptic plasticity as the fundamental associative mechanism and extends to the development of "cell assemblies" that represent specific percepts, thoughts, and actions, in all species including humans. Modern neuroscience since Hebb's time has made tremendous advances in understanding synaptic plasticity mechanisms and their possible role in memory using mammalian model systems. This chapter reviews some of the recent progress toward a full characterization of one particular form of synaptic plasticity observed in the mammalian brain called long-term potentiation (LTP). LTP is a laboratory phenomenon, but its mechanisms are now quite well understood. It can be induced in many brain structures that are involved in memory, and there is substantial evidence that the same cellular mechanisms that mediate LTP are required for lasting memory. Therefore, this chapter reviews the state of our understanding of this important phenomenon as a likely candidate for memory coding in mammalian systems. Hippocampal long-term potentiation as a model memory mechanism Long-term potentiation is most commonly studied in the hippocampus, a brain structure that you will come to know quite well in this book. The hippocampus is a complex structure, but it is easy to find in the brain, and its inputs, outputs, and intermediate pathways are largely segregated, making it an excellent model system for studying its circuitry. We now know a lot about the initial steps in the molecular and synaptic basis of LTP, particularly as seen in the hippocampus. The elucidation of these mechanisms has been facilitated greatly by the development of the in vitro hippocampal "slice" preparation in which thick transverse sections of the hippocampus are taken from the brain and kept alive in a Petri dish (Fig. 3-1). This preparation lacks the complex influences of the normal inputs and outputs of the hippocampus, but provides an especially clear access to cells and intrinsic connections of the hippocampal circuit. Most of these studies have focused on area CA1 of the hippocampus, where the in vitro preparation allows multiple input and output pathways to be preserved intact and to be manipulated independently for recording and stimulation (Fig. 3-2A).

Cellular Mechanisms of Memory: Complex Circuits

55

Figure 3-1. The hippocampal slice preparation. A: A view of the rodent brain. The hippocampus is a large structure inside the cerebral hemispheres. A slice is taken in a plane transverse to the long axis of the hippocampus. B: The hippocampal slice in vitro, with indications of its major subdivisions, CA1, CA3, DG = dentate gyrus, EC = entorhinal cortex.

The phenomenon of LTP was first discovered by Terje Lomo, a PhD student working in Oslo. Lomo was exploring the physiology of the circuitry of the hippocampus, and in particular he was examining the phenomenon of frequency potentiation, an increase in the magnitude of responsiveness of cells following a series of rapidly applied activations. Lomo observed that repetitive high-frequency electrical stimulation (called tetanus) of one pathway resulted in a steeper rise time (slope) of the excitatory synaptic potential to a subsequent single pulse. He also observed that following a tetanus there was recruitment of a greater number of cells reaching the threshold for an action potential, reflected in a greater "population spike," the spike observed when many cells fire together (see examples in Fig. 3-2B). Lomo found that the tetanus-induced changes in the synaptic and cellular responses to single pulses lasted for several hours,

56

Part I: Connection

Figure 3-2. Hippocampal long-term potentiation (LTP). A: Illustration of a horizontal section through the hippocampus showing the pathways by which pyramidal cells in CA1 are stimulated by either a strong input from CA3 or a weak input from the entorhinal cortex (EC). B: Excitatory postsynaptic potentials (EPSPs) recorded after single pulse stimulations of the strong path before (left) and after (right) tetanus. Below is the standard method for tracking the changes in the EPSP slope over a period of hours. C: Associative LTP. A schematic diagram of a CA1 pyramidical cell and loci of strong and weak stimulation, and measurements of the EPSP slopes following single pulses of each type of input. The weak input stimulation alone (open arrow at 0.5 hr on lower graph) produces only a transient change, but strong stimulation alone (filled arrow at 1 hr) produces LTP. Combined strong and weak stimulation (both arrows at 1.5 hr) result in LTP at both synaptic sites (data from Bliss and Collingridge, 1993).

leading him to distinguish this phenomenon from short-lasting facilitations. And so, he called it "long-term potentiation." In subsequent years several investigators have characterized the basic properties of the synaptic and cellular components of LTP, creating considerable excitement about this phenomenon as a model for lasting history-dependent synaptic change. The findings have spawned a veritable cottage industry within the field of neuroscience.

Cellular Mechanisms of Memory: Complex Circuits

57

What fascinated researchers at the outset were the remarkable parallels between properties of LTP and memory. In 1989 Richard Morris identified five fundamental properties that make LTP such an attractive model of memory. First, LTP is a prominent feature of the physiology of the hippocampus, a brain structure universally identified with memory. Subsequent work has made clear that the hippocampus is not the only site of LTP, but its functional role as a component of one of the brain's major memory systems would seem to demand that it possess a memory mechanism. The second and third properties have to do with temporal characteristics. LTP develops very rapidly, as one would require of a plausible memory mechanism, typically within 1 minute after a single stimulus train delivered with the proper parameters. Moreover, like a good memory, LTP can be long-lasting. In in vivo preparations it can be observed for hours after a single stimulation train, or for weeks or more after repetitive stimulations that might act as "reminders." Fourth, LTP has the sort of specificity one would require of a memory mechanism: Only those synapses activated during the stimulation train are potentiated. Other neighboring synapses, even on the same neurons, are not altered. This phenomenon parallels the natural specificity of our memories, in which we are able to remember many different specific episodes with the same person (e.g., one particular date you had, out of many, with a given individual) or object (e.g., where you parked your car today rather than last week), and thus would be a key requirement of any useful cellular memory mechanism. In addition, the property of specificity may be key to the magnitude of the storage capacity of brain structures. Each cell can participate in the representation of multiple memories, each composed of distinct subsets of its many synaptic inputs. Fifth, and perhaps most definitively important for memory, LTP is associative in that potentiation occurs best when multiple inputs are stimulated simultaneously during the tetanus (Fig. 3-2C). This phenomenon has been demonstrated most elegantly in studies that employ activation of separate pathways that synapse on the same hippocampal neurons. In these studies the two pathways involve the combination of a "weak" input, designated as one that does not produce potentiation at any stimulation level, plus a "strong" input for which a threshold level of stimulation suffices to produce LTP. Associativity is observed when the weak input is activated at the same time as the strong input, resulting in LTP of the weak as well as the strong pathway. The time window for this sort of association was initially thought to be quite brief, on the order of a few milliseconds, and thus quite limited in the extent to which it could support Pavlovian condition-

58

Part I: Connection

ing, which usually involves hundreds of milliseconds separation between conditioned stimulus (CS) and unconditioned stimulus (US). However, there is new evidence that a form of associativity may be possible within a broader, and more behaviorally meaningful, time window. An intriguing 1997 study by Frey and Morris indicated that activity at hippocampal synapses that produces only a short-lived potentiation can nonetheless create a synaptic "tag" that lasts a few hours. Subsequent strong activation of a neighboring pathway within that period leads to lasting potentiation of both the "strong" pathway and the previously "tagged" synapses. Thus, LTP could serve to associate or integrate patterns of activity over a time window that has obvious behavioral significance. The property of associativity is especially appealing because it offers a cellular model of the mechanism for structural change in neural connections, a change that would increase synaptic "efficacy," the strength of the postsynaptic response resulting from a presynaptic activation. When Hebb proposed his theory of cell assemblies (see Part III), he recognized the need for alterations in the cells so that a whole assembly could be reactivated to recall a memory. He suggested that the essential trigger for changing synaptic efficacy involved the repeated activation of a presynaptic element AND its participation in the success in firing the postsynaptic cell. The simultaneous activation of many inputs of the hippocampus during a tetanus provides a perfect situation to accomplish the co-occurrence of presynaptic and postsynaptic activity. Furthermore, the property of associativity, by permitting the ability to integrate patterns of activity, simultaneously satisfies the induction requirement of LTP—that there be a combination of presynaptic and postsynaptic activation—and offers a fundamental mechanism for encoding associations between functionally meaningful activation patterns. Cellular basis for the induction of hippocampal LTP Many studies have shown that the induction of LTP in area CA1 requires two fundamental synaptic events—activation of presynaptic inputs and depolarization of the postsynaptic cell. Both are ordinarily accomplished within a single high-frequency stimulus train—the initial stimulation depolarizes the cell for a relatively prolonged period during which the following stimulations provide simultaneous postsynaptic activations. However, high-frequency stimulation is not required per se. Instead, for example, direct depolarization of the postsynaptic cell by injection of current through an intracellular electrode, combined with low-frequency presynaptic input, will suffice. Conversely LTP induction can be blocked

Cellular Mechanisms of Memory: Complex Circuits

59

by preventing depolarization, or by hyperpolarization of the postsynaptic cell. These findings show that the conditions of Hebb's postulate about the critical conditions for cellular events are both necessary and sufficient to provide a synaptic mechanism for memory. Molecular bases for the induction and maintenance of hippocampal LTP The molecular mechanism that underlies the major type of LTP induction in CA1 involves special properties of a combination of synaptic receptors, some of which should be familiar from your reading of Chapter 2 (see Fig. 3-2). Considerable evidence points to the amino acid glutamate as the primary excitatory transmitter in the hippocampus and elsewhere where LTP is found. There are several types of glutamate receptors, most prominently divided into those that are excited by N-methyl-D-aspartate (NMDA receptors, already introduced in Chapter 2) and those that are activated by a-amino-3-hydroxy-5-methyl-4-isoxazolepropionate (AMPA receptors). These two types of receptors can be dissociated functionally by pharmacological manipulations. In particular, the NMDA receptors are selectively and competitively blocked by the antagonist D-2-amino-5-phosphonovalerate (AP5). A major discovery in revealing the mechanism of LTP was that AP5 has little effect on excitatory postsynaptic potentials (EPSPs) elicited by low-frequency stimulation, indicating that AMPA receptors, and not NMDA receptors, mediate normal synaptic transmission in the hippocampus. In contrast, AP5 completely blocks LTP following highfrequency stimulation trains, indicating that glutamate activation of NMDA receptors is critical to this form of synaptic plasticity. Discoveries about two major differences between NMDA and AMPA receptors in their regulation of postsynaptic ion permeability in CA1 offer an explanation of the role of these receptors in LTP (Fig. 3-3). First, activation of AMPA receptors increases the permeability of the postsynaptic membrane to both sodium (Na + ) and potassium (K + ) ions, but does not alter cell permeability to calcium (Ca 2+ ). By contrast, activation of NMDA receptors increases permeability to Ca2+ as well as to Na+ and K+ ions. Second, unlike for AMPA receptors, ion flow through NMDA receptors is highly dependent on the voltage state of the postsynaptic cell at the time of NMDA receptor activation. In the resting state, NMDA receptor channels are blocked by another doubly charged ion, magnesium (Mg2+), which prevents the flow of the other ions even in the presence of glutamate at the receptor. However, when the membrane of the postsynaptic cell is depolarized, Mg2+ is expelled from the receptor channel, al-

60

Part I: Connection

Figure 3-3. Molecular mechanism of the induction of long-term potentiation (LTP). GLU = glutamate; See text for explanation (adapted from Nicoll et al., 1988).

lowing glutamate to bind and the ions including Ca2+ to flow. Thus, the effect of the initial activations in the high-frequency stimulus train is to activate the AMPA receptors, depolarizing the postsynaptic cell membrane. This unblocks the NMDA receptor channels so that succeeding stimuli activate the NMDA receptor, allowing Ca2+ to enter the postsynaptic cell.

Cellular Mechanisms of Memory: Complex Circuits

61

The entry of Ca2+ to the intracellular space is a key step in the induction of LTP. This is shown in at least three lines of evidence. First, LTP is prevented when Ca2+ is bound by intracellular injection of a calcium chelator (a molecule that binds up calcium). Second, LTP is triggered by intracellular injection of a caged Ca2+ compound that releases calcium molecules. Third, the entry of Ca2+ into the postsynaptic cell following stimulation trains has been directly imaged using a sophisticated technique called confocal fluorescence microscopy. There is evidence, however, that Ca2+ entry does not, by itself, lead to lasting synaptic potentiation; rather, some sort of NMDA receptor activation seems to be required. One possibility under scrutiny is that glutamate also activates metabotropic receptors that mediate release of intracellular stores of Ca2+ as an amplification mechanism. The succeeding steps in the permanent maintenance of LTP are less well understood, and can only be provided in outline form at this time (Fig. 3-4). The leading view is that the role of Ca2+ is to activate kinases, enzymes that phosphorylate proteins, transforming them into their active

Figure 3-4. Model of the molecular mechanisms of short-term and long-term processes following long-term potentiation induction.

62

Part I: Connection

configuration, and some of these will be familiar from your reading about invertebrate systems in Chapter 2. Specific candidates of the critical kinases in the hippocampus include type II Ca2/calmodulin-dependent kinase (CaMKII), Ca2+/phospholipid-dependent protein kinase C (PKC), and mitogen-activated protein kinase (MAPK). CaMKII is a very attractive candidate because it is present in large quantities in the postsynaptic area and its initial activation depends on Ca2+. Tetanus stimulates its production, and pharmacological inhibition and genetic elimination of CaMKII block LTP. Furthermore, following activation, CaMKII undergoes an autophosphorylation by which it becomes independent of the transient Ca2+ influx to remain phosphorylated. This prolonged activation could mediate long-lasting consequences, one of which might be the conversion of inactive (so-called silent) AMPA receptors into active ones. PKC is also strongly implicated by experiments showing that its activation results in marked potentiation of the EPSP that occludes further potentiation by stimulation trains. Furthermore, intracellular injection of PKC enhances synaptic transmission, and application of PKC antagonists block LTP. MAPK is activated by phosphorylation following LTP or stimulation that results in intracellular Ca2+, and inhibition of MAPK prevents later steps in gene expression. In addition to modifying existing proteins, there is evidence that the maintenance of LTP also depends on new protein synthesis in the hippocampus. Experiments using protein synthesis inhibitors indicate that proteins synthesized from preexisting mRNA are required for lasting LTP. In addition, there is also evidence that the maintenance of LTP depends upon the cAMP-responsive transcription factor CREB. One possible mechanism for CREB involves the Ca2+ influx activating adenylyl cyclase, which in turn activates cAMP. This could in turn activate PKC, leading to the phosphorylation of many proteins, including CREB. The phosphorylated form of CREB is known to modulate the transcription of genes so as to increase the expression of several proteins. There is some evidence indicating that genetically altered mice who lack a form of CREB have deficient LTP maintenance. One possible target of new protein synthesis in the hippocampus is the production of neurotrophins, molecules long known as regulated by neural activity and having the capacity to promote morphological change and increased connectivity. Stimulation trains capable of inducing LTP increase the gene expression for some neurotrophins in the hippocampus. In turn, some neurotrophins potentiate glutaminergic transmission in CA1, and these effects occur within minutes and last for several hours or longer.

Cellular Mechanisms of Memory: Complex Circuits

63

Where is the synaptic alteration? A major unresolved question is whether the locus of lasting synaptic alteration following LTP is presynaptic or postsynaptic. Of course, changes at both sites are entirely possible. The evidence on both sides of this issue is considerable, and it may not be possible to fully resolve the issue with current methods available to study the hippocampal preparation. Recent attempts to resolve the question have focused on quantal analysis, a protocol that involves reducing presynaptic release to a statistical phenomenon. This method allows estimation of the magnitude of the postsynaptic response to a single quantum of transmitter, presumably corresponding to release of a single synaptic vesicle. During LTP there is a decrease in the percentage of failures of postsynaptic response and a decreased variability of responses to presynaptic stimulation, both consistent with an increase in the probability of presynaptic release. However, there is also observed an increase in the amplitude of the response to a single quantum of transmitter, which is consistent with an increase in postsynaptic receptor efficacy. Thus, the current evidence from quantal analyses are consistent with both loci as being involved in plastic change. Conceptual considerations have also weighed in on this controversy. Possible cellular mechanisms for postsynaptic modification are straightforward to envision, as just discussed. By contrast, because the initial effects of combined pre- and postsynaptic activity evoke cellular mechanisms localized in the postsynaptic cell, an ultimate change in presynaptic physiology would require production and transport of some sort of retrograde messenger, a signal that travels from the activated postsynaptic site to the presynaptic site. Several candidates for the retrograde messenger have been proposed [e.g., arachidonic acid, nitrous oxide (NO), carbon dioxide (CO)], but so far, none has received more than fragmentary support.

Hippocampal long-term depression If there was only a form of plasticity that increased synaptic efficacy, eventually all synapses would become "saturated," that is, raised to a ceiling level of efficacy, and no further learning could occur. So most investigators think that, in addition to the potentiation of synapses, there must be a mechanism of depotentiation or long-term depression (LTD) of synaptic efficacy. LTD also can enhance the relative effect of LTP at neighboring synapses, improving the signal-to-noise contrasts, as well as also increasing the range of synaptic coding patterns by a population of synapses providing input to a single postsynaptic cell.

64

Part I: Connection

In general, the learning rule for LTD involves activity-dependent plasticity with a direct violation of the Hebb rule for LTP. Thus, LTD has been described under conditions where there is either presynaptic activity or postsynaptic activity, but not both. One can activate presynaptic elements with single pulses at a very low rate that produces no activation, or only weak activation, of the postsynaptic cell. Or one can induce presynaptic activity in the absence of postsynaptic firing, by activating presynaptic elements weakly and out-of-phase with strong stimulation to converging synapses of the same postsynaptic cell. Alternatively, LTD also results from activation of the postsynaptic neuron, without activation of the presynaptic element. This form of LTD can be induced either by stimulation of separate converging synapses on the same postsynaptic cell or by inducing an action potential in the axon that is conducted backward to the cell body (called antidromic activation). All of these forms of LTD have been observed at one pathway or another in the hippocampus, but it remains to be seen if they obey the same induction and maintenance rules and are available at all sites in the hippocampus. Anatomical modifications consequent to LTP Most researchers believe that lasting changes in neural connectivity ultimately require altered morphology of synapses. Although this research area has been plagued by technical issues of the proper means of preserving tissue for examination with the electron microscope, there is now substantial evidence that LTP does result in structural alterations in synaptic connections consistent with increases in synaptic efficacy. Most of these data focus on the protruding heads of dendritic spines that are the excitatory synapses of the dentate granule or CAl pyramidal cells, the same sites that involve NMDA receptor dependent LTP. In the dentate gyms, the reported structural changes suggest an increase in spine surface area and in the area of the opposing pre- and postsynaptic membranes. In some reports the changes in spines occurred without any appreciable increase in the number of synapses, suggesting an interconversion of spine shapes in which LTP results in increases in synaptic contact area by expansions of the preand postsynaptic cell membranes surrounding one another. Studies on CAl have provided strikingly parallel results, changes in spine dimensions consistent with the overall rounding of spines, although these changes may be transient. Recent studies using newly available high resolution optical methods have detected growth of new spines on postsynaptic dendrites in CAl shortly after induction of LTP. In addition, lasting changes in spine number have been reported in CAl, and these changes have been charac-

Cellular Mechanisms of Memory: Complex Circuits

65

terized as reflecting the transformation of synapses into types with protracted necks as well as sessile types. The coordination of increased presynaptic and postsynaptic active contact, as observed in studies on both the dentate granule and CA1 pyramidal cells, seems to obviate the question of whether the fundamental basis of LTP is pre- or postsynaptic in origin. LTP beyond the hippocampus

LTP was first discovered in the hippocampus, and it is easily studied there because of the laminar separation of synaptic inputs and outputs. However, there are now reports of potentiation of synaptic efficacy in widespread areas of the brain and LTP is rapidly becoming viewed as a universal plasticity mechanism. Among the areas where LTP, and/or LTD, have been demonstrated are several areas of the neocortex, piriform cortex, amygdala, striatum, cerebellum, and even the spinal cord. Perhaps best characterized of the nonhippocampal areas is the visual cortex, which has been studied extensively by Mark Bear and his colleagues. In the rat visual cortex, Bear's research group developed an in vitro visual cortex slice preparation in which they would stimulate layer IV input cells and record from layer III principal cells that receive inputs from the layer IV cells (Fig. 3-5). They recorded EPSPs before and after

Figure 3-5. Induction of long-term potentiation (LTP) and long-term depression (LTD) in hippocampal and cortical slices. The plots show that in both preparations an increase in the excitatory postsynaptic potentials (EPSP) (LTP) is produced by short bursts of high-frequency stimulation (HFS), and in both preparations a decrease in the field EPSP (LTD) is produced by low-frequency stimulation (LFS) (data from Bear, 1996).

66

Part I: Connection

tetanization of the input cells, and found that stimulation of the input layer results in LTP and LTD in principal cells with the same protocols effective for producing these phenomena in the hippocampus. Thus, they demonstrated bidirectional activity-dependent modification of visual cortex synapses such that low-frequency stimulation results in synaptic depression, whereas high-frequency stimulation produces potentiation, just as in the hippocampus. Both forms of synaptic modification are synapse specific and depend on NMDA receptors. Intracellular injections of current that produce postsynaptic depolarization or hyperpolarization paired with low-frequency synaptic activation produce synaptic enhancements or decrements, respectively. LTP and memory LTP captures an exciting physiological phenomenon, one that is seen, deservedly, as the most prominent model of synaptic plasticity that might underlie memory. As Charles Stevens once put it, this mechanism is so attractive that it would be a shame if the mechanism underlying LTP turned out not to be a memory mechanism. But there should be no doubt about the fact that LTP is not memory—it is a laboratory phenomenon that involves massive coactivations never observed in nature. The best we can hope is that LTP and memory share a common mechanism. In recent years disappointing evidence has emerged, amidst the more positive findings, regarding all main lines of evidence that have been offered to connect LTP and memory. Here I attempt just to summarize the history of the research on the possible linkage between LTP and memory. Several relatively direct approaches have been pursued in attempting to demonstrate that LTP and memory share common physiological and molecular bases. Most prominent are demonstrations of changes in synaptic efficacy consequent to a learning experience ("behavioral LTP"), and, conversely, attempts to prevent learning by pharmacological or genetic manipulation of the molecular mechanisms of LTP induction. Examples of each approach are presented here, and discussed in light of the inherent limitations they have in convincingly connecting LTP and memory. "Behavioral LTP"

Do conventional learning experiences produce changes in synaptic physiology similar to the increases in EPSP and cellular responses that occur after LTP? Seeking changes in synaptic physiology consequent to learning is an ambitious and optimistic approach because one might well expect the

Cellular Mechanisms of Memory: Complex Circuits

67

magnitude of synaptic change observed in gross field potentials to be vanishingly small following any normal learning experience—a virtual "needle in a haystack." In addition, it is most likely that learning involves changes in synaptic efficacy in both the positive and negative directions, that is, both LTP and LTD. Thus, learning would likely result in changes in the distribution of potentiated and depressed synapses with little or no overall shift, and consequently no change, or even an overall negative change, in the averaged evoked potentials commonly used to measure LTP. Addressing the first of these concerns by using powerful and extended experience as the learning event, the initial reports showed enhancement of excitatory synaptic potentials and population spikes after different types of learning experience. The early studies include ones in which several aspects of synaptic physiology were observed to change in the perforant pathway response in rats who had been exposed for prolonged periods to an "enriched" as compared to "impoverished" environment. The "enriched" rats lived in a large housing area with littermates, with continuous social stimulation and various forms of environmental stimulation through their opportunity to investigate and interact with many objects placed in their shared cages, whereas the "impoverished" rats had solitary housing, the absence of stimulating objects, and a small living space. In one particularly illustrative study of this type, hippocampal slices were taken from these rats and were tested for various aspects of synaptic and cellular responsiveness. It was found that rats who had lived in the enriched environment, compared to those restricted to the impoverished environment, had an increased slope of the synaptic potential and larger population action potentials, implying more cells recruited, but no change in other physiological parameters. These changes are entirely consistent with the pattern of increased synaptic efficacy observed following LTP. These changes were not permanent, however. They disappeared if the enriched-condition animals were subsequently isolated for 3-4 weeks prior to the analyses. Recent observations by Joseph LeDoux and his colleagues offer confirming evidence for a connection between the phenomena of LTP and enhanced transmission of relevant sensory inputs in a different neural circuit that supports a specific kind of learning. In this case the learning involved a form of classical (Pavlovian) conditioning in which rats become fearful of tones that have been paired with foot shocks (see Chapter 11). The relevant anatomical pathway involves auditory inputs to a subcortical structure in the thalamus called the medial geniculate nucleus, projections from there to another subcortical area called the lateral amygdala nucleus, and then projections from there to other parts of the amygdala which control the expression of fear responses (Fig. 3-6A; this pathway is described in

68

Part I: Connection

Figure 3-6. Common mechanisms for long-term potentiation (LTP) and sensory processing. A: Anatomical pathway for natural auditory stimulation or electrical stimulation of the medial geniculate nucleus and recording in the lateral amygdala. B: Examples of excitatory postsynaptic potentials (EPSPs) evoked by auditory stimulation or electrical stimulation of the medial geniculate before and following highor low-frequency medial geniculate electrical stimulation that produces LTP in the lateral amygdala recording; note that the same stimulation at low frequency does not produce any potentiation of the EPSP. The induction of LTP in that pathway also results in a large increase in the EPSP response to auditory stimulation in the same pathway (data from Rogan and LeDoux, 1995).

more detail in Chapter 11). In one study they used high-frequency electrical stimulation of the medial geniculate nucleus to induce LTP within the lateral nucleus of the amygdala. Consequently they found an enhancement of synaptic responses within the same area of the amygdala to natural auditory stimulation (Fig. 3-6B). In a complementary study they found the converse evidence that fear conditioning enhances early sensoryevoked responses of neurons in the lateral amygdala. These findings are illuminating in two ways. First, they support the view that there is nothing special about the hippocampus when it comes to LTP. Second, the approach taken by LeDoux and colleagues points toward a potentially more decisive and therefore possibly more fruitful way to link LTP and memory. The pattern of changes in auditory-evoked synaptic potentials, including the magnitude, direction, and longevity of increased responses, paralleled those parameters for electrically induced synaptic potential, showing us that natural information processing can make use of the very cellular and molecular mechanisms set in place by conventional, artificially induced LTP. Conversely, the observation of in-

Cellular Mechanisms of Memory: Complex Circuits

69

creased neuronal sensory responses following conditioning shows us that, like LTP, real learning can enhance information processing relevant to the task. In subsequent studies the same group has also now shown that repeated pairings of auditory stimuli and foot shocks that train rats to fear the tones also alter evoked sensory responses to the tones in the same way as LTP in that pathway (Fig. 3-7). Thus, in rats with properly timed pairings, tones produce evoked potentials of greater slope and amplitude, just as electrical stimulus trains do when applied to this pathway. No enhancement of field potentials is observed with unpaired tone and foot shock presentations, even though this conditioning control leads to as much of a behavioral response (freezing) as paired presentations (even the unpaired control rats learn to freeze to the environmental context where shocks are received—see Chapter 11). Furthermore, this behavioral LTP is enduring, lasting at least a few days, as long as the behavioral response during extinction trials. Thus, LeDoux and colleagues' approach takes us beyond mere similarities between LTP and memory, bringing into contiguity the identical neural pathways and experimental procedures that define LTP and sensory processing in memory. A different set of studies on the rat motor cortex by John Donoghue and his colleagues demonstrates the generality of this approach to other brain areas and other forms of learning. In these experiments, rats were trained to reach with one particular paw through a small hole in a food

Figure 3-7. Changes in auditory field potentials in the amygdala following fear conditioning. Note the sustained increase in the size of the auditory-stimulation-induced excitatory postsynaptic potentials (EPSP) following paired tone-shock combinations versus no change in the EPSP size under control conditions where tones and shocks were not consistently paired (data from Rogan et al., 1997).

70

Part I: Connection

box in order to retrieve food pellets. Initially, the rat's reaching movements are labored and often the food pellets are dropped. However, over the course of one or two hour-long practice sessions, the rats refine their motor coordination and ultimately obtain pellets at a rapid asymptotic rate. Following this kind of motor coordination training, Donoghue and colleagues removed the brain and measured the strength of connections among cells within the area of the motor cortex that controls hand movements. They accomplished this using an in vitro preparation of the appropriate brain section and evoking EPSPs in a principal cell layer of the motor cortex by stimulating horizontal fibers that connect neighboring cells to one another (Fig. 3-8). They found that for the same or lower input stimulation intensity, the magnitude of the EPSPs on the side of the brain that controlled the trained paw (i.e., in the contralateral or opposite hemisphere) were consistently larger than those on the side of the brain that controlled the untrained paw. Furthermore, they also found it difficult to induce LTP by electrical stimulation in the trained hemisphere, but not in the untrained hemisphere. Thus, training produced an anatomically localized increase in synaptic efficacy that occluded the capacity for LTP. These observations show in a compelling way that synaptic potentiation results from motor learning, and the real plasticity phenomenon shares common resources with the artificial one, providing strong evidence for common cellular mechanisms of LTP and learning. Blocking LTP and memory

The major limitation of the preceding approach is that the experiments only provide correlations between aspects of LTP and memory. The con-

Figure 3-8. Effects of motor skill learning on evoked excitatory postsynaptic potentials (EPSPs) in the motor cortex. A: Placement of stimulating and recording electrodes in the two sides of the motor cortex slices. B: (example) and C: (group average) EPSPs recorded from trained animals and untrained controls. The dark lines represent recordings from the trained hemisphere and dotted lines recordings from the untrained hemisphere (from Rioult-Pedotti et al., 1998).

Cellular Mechanisms of Memory: Complex Circuits

71

verse approach is to draw cause-and-effect links between the phenomenology of LTP and of memory by blocking LTP and determining if memory is prevented. Perhaps the most compelling and straightforward data on a potential connection between the molecular basis of LTP and memory have come from experiments where a drug or genetic manipulation is used to block LTP and, correspondingly, prevent learning. Here again there was the need for optimistic assumptions. It had to be assumed that the drugs were selective to plasticity and not normal information processing in the brain, and that they would knock out a critical kind of plasticity. These assumptions were accepted based on the observation that drugs such as AP5, which selectively blocks the NMDA receptor, prevent hippocampal LTP while sparing normal synaptic transmission. Thus, to the extent that the role of the NMDA receptor is fully selective to plasticity, one might predict these drugs would indeed block new learning without affecting nonlearning performance or retention of learning normally accomplished prior to drug treatment. Consistent with these predictions some of the earliest and strongest evidence supporting a connection between LTP and memory came from studies on spatial learning by Richard Morris and his colleagues. Morris developed a maze learning task in which hippocampal function is required. In this task, the maze involves a swimming pool in which the water is made opaque by the addition of a milky powder, and an escape platform is submerged just under the water at a predetermined location. Rats are good swimmers, but prefer to find and climb onto the platform. Typically, they are trained to find the platform from any of four starting positions around the periphery of the maze, and they show learning by shortening the time required to escape from all starting points. At the end of training, their memory is assessed using a probe test where the platform is removed, and rats exhibit good memory by swimming in the vicinity of the former escape locus. Initially Morris and his colleagues showed that AP5-induced blockade of NMDA receptors prevents new spatial learning in the water maze. Drugtreated rats swim normally, but do not reach the same level of rapid escape as normal rats. Indeed, the drug-treated rats often adopt a strategy of swimming at a particular distance from the walls of the maze, which reduces their escape latency to some extent without knowing the exact location of the platform. In the probe tests, normal rats show a distinct preference for swimming in the vicinity of the former escape locus, but drugtreated rats show little or no such bias, indicating the absence of memory for the escape location (see Chapter 5 for further discussion of this task). Additional experiments showed no effect of AP5 on retention when train-

72

Part I: Connection

ing was accomplished prior to drug treatment. This could be fully predicted because NMDA receptors are viewed as required only for the induction of LTP and not for its maintenance (see Molecular bases for induction and maintenance of hippocampal LTP). Furthermore, the deficit was limited to spatial learning, known from other work to be dependent on the hippocampus, and not to a simple visual discrimination, known to be NOT dependent on hippocampal function. Additional research by Morris and his colleagues has also shown how NMDA receptor dependent LTP might play a continuing role in updating one's memories. To accomplish this they developed a new version of the water maze task in which the location of the escape platform is moved every day, and animals are given four trials to learn the new location. Across a series of training days the rats became skilled at the task such that they consistently found the platform very rapidly on the second trial it was presented. Subsequent to initial drug-free training, animals were tested with different memory delays inserted between the first and second trial on each day (Fig. 3-9A). On some days, AP5 was infused into the hippocampus, and on other days a placebo was given. AP5 treatment resulted in a deficit on trial 2 performance (Fig. 3-9B). Moreover, this deficit was dependent on the time interval between trial 1 and trial 2, such that no impairment was observed with a 15 second intertrial interval, but significant deficits ensued if the intertrial interval was extended to 20 minutes or longer. These data suggest that memory for specific episodes of spatial learning remains dependent on NMDA receptors and LTP, even after the animals have learned the environment and the general rules of the spatial task. Genetic manipulations of LTP and memory

Other research has used targeted genetic manipulations to show that blocking the cascade of molecular triggers for LTP also results in severe memory impairments. In one of the early studies of this type, mice with a mutation of one form of CaMKII had deficient LTP and were selectively impaired in learning the Morris water maze. Despite the fact that the genetic manipulation was effective throughout development, the hippocampus appeared normal in architecture and was normal in its basic physiological responsiveness. Since that time, selective memory impairments have been reported in several different types of knockout mice with deficiencies in LTP, with a special emphasis on knockouts of CREB. The manipulation of biochemical mechanisms by interference with specific genes allows investigators to identify critical molecular events at a very high level of specificity. For example, one study by Alcino Silva and

Cellular Mechanisms of Memory: Complex Circuits

73

Figure 3-9. Matching-to-place version of the Morris water maze task A: Example test sessions. On each session the animal is given four trials with the escape platform in a novel location. Note that the first intertrial interval is variable, and the others are constant at 15 sec. B: Effects of infusion of AP5. After initial training on the task, rats treated with AP5 or placebo were tested on a series of four-trial sessions with the escape platform in a novel location each day. On some test sessions the interval between trials 1 and 2 was 15 sec. On these sessions rats given AP5 performed as well as normal subjects in showing a substantial reduction in the latency that indicated intact memory On other sessions the interval between trials 1 and 2 was 20 min or 2 hr. On these sessions normal animals also showed good retention, but AP5 rats showed substantially less reduction in their escape latencies, indicating memory impairment. The later intertrial intervals were all 15 sec, and all animals show substantial latency decreases over these brief intervals. Filled circles = controls; open circles = AP5 (data from Morris and Frey, 1997).

his colleagues showed that substitution of a single amino acid in CaMKII that prevents its autophosphorylation results in severe learning and memory deficits. In addition, other new genetic approaches are providing greater temporal- as well as region-specific blockade of gene activation. In another recent study by Susumu Tonegawa and his colleagues, the genetic block was limited to postdevelopment activation of the genes for the NMDA receptor specifically in the CA1 subfield of the hippocampus,

74

Part I: Connection

which selectively blocked LTP in that region. Despite these highly selective temporal and anatomical restrictions, the mice with this mutation were severely deficient in spatial learning as well as in other types of memory dependent on hippocampal function. A complementary recent study showed that a mutation that results in overexpression of NMDA receptors can enhance several kinds of memory dependent on the hippocampus. There is also growing evidence that interference with other events involved in the LTP molecular cascade, specifically PKC and MAPK, also have deleterious effects on memory. Thus, it is likely that the full set of cellular events that mediate LTP will also be shown to play critical roles in memory. Blocking LTP and plasticity of hippocampal firing patterns The studies described previously provide strong evidence in favor of the view that pharmacological and genetic manipulations that prevent hippocampal plasticity selectively block a critical stage in memory formation. However, skeptical neuroscientists are always concerned that a particular drug or genetic manipulation could have its deleterious effects not on memory directly, but rather on the normal information processing in brain structures that play a role in task performance. In general the best evidence offered in pharmacological and genetic studies involves the observation that the drugs or genetic manipulations do not affect synaptic transmission as revealed in evoked potential protocols. However, it is important to realize that large-scale evoked EPSPs never actually occur during normal information processing. So the data available from these studies do not allow us to conclude that other more complex, and more relevant, patterns of hippocampal information processing are fully normal under the influence of drugs such as AP5. A newer generation of combined electrophysiological and pharmacological-genetic studies is providing evidence critical to this question. Several studies have how examined the nature and persistence of spatial representations of single hippocampal neurons and neuronal populations in animals with compromised capacity for LTP. These studies involve genetically altered mice or rats with pharmacologically blocked LTP and recordings of so-called place cells, hippocampal neurons that fire when the animal is in a particular location in its environment (see Chapter 6 for a detailed explanation of place cells). One of these studies found that mutant mice expressing an active form of CaMKII that impairs one form of LTP and spatial learning have impoverished and unstable hippocampal spatial representations. Hippocampal cells of these mice initially develop

Cellular Mechanisms of Memory: Complex Circuits

75

spatially specific firing patterns, albeit in fewer cells, and the spatial specificity of these patterns is reduced. Perhaps most important, unlike normal mice who have very stable hippocampal spatial representations, the spatial firing patterns in mutant mice are lost or changed if the animal is removed from the environment and later replaced even within a few minutes. Acute pharmacological blockade of NMDA receptors in rats also resulted in instability of the spatial firing patterns of hippocampal neurons, without affecting the incidence or spatial specificity of previously acquired spatial firing patterns. The drug did not prevent the initial establishment of hippocampal spatial firing patterns or their short-term retention between repeated recording sessions separated by brief intervals. By contrast the maintenance of a newly developed spatial representation across days was severely compromised. The consequence of LTP blockade for the network processing of hippocampal spatial representations has also been examined. One study examined the spatial firing patterns of groups of neighboring hippocampal neurons in mice with the CA1-specific knockout of NMDA receptors. They also reported that these cells had diminished spatial specificity, and characterized a reduction in the coordinated activity of neurons tuned to overlapping spatial locations. Furthermore, they tied these findings to the spatial memory impairment by showing how the loss of coordinated activity in mutant hippocampal place cells leads to a poorer prediction of sequential locations during navigation behavior. Another study characterized hippocampal spatial representations in mice with knockouts of CaMKII or CREB. Similar to the other studies, they observed diminished spatial selectivity in both mutants, as well as diminished stability of the spatial representations when some of the environmental cues were altered. CaMKII mutant mice could not recover their spatial representations when the environmental cues were returned to their original configuration. However, the CREB-knockout mice, in whom spatial learning and LTP are partially preserved, showed they could recover their spatial representations in the original environment. These results suggest that the network processes that bind together single neuron representations of spatial cues are particularly dependent on LTP. Further investigations on the coding of space by hippocampal networks offer a particularly promising direction for relating synaptic plasticity processes to memory functions. Blocking LTP and memory outside the hippocampus Other studies suggest that the cascade of molecular events that is invoked by LTP may also mediate cortical plasticity that underlies memory. A par-

76

Part I: Connection

ticularly good example of this work involves a set of studies by Yadin Dudai and his colleagues focused on taste learning mediated by the gustatory cortex of rats. When rats are exposed to a novel taste and subsequently become ill, they develop a conditioned aversion specifically to that taste, and this learning is known to depend on the gustatory cortex. Blockade of NMDA receptors by infusion of the antagonist AP5 produces an impairment in taste aversion learning, whereas the same injections given prior to retention testing, or into an adjacent cortical area, had no effect. Thus, it is likely that modifications in cortical taste representations depend on LTP. Furthermore, blockade of protein synthesis in the gustatory cortex by infusion of an inhibitor prior to learning also prevents development of the conditioned taste aversion. By contrast, the same injection given into a neighboring cortical area or given to the gustatory cortex hours after learning has no effect. Consistent with this finding, MAP kinase as well as a downstream protein kinase were activated selectively in gustatory cortex within 10 minutes of exposure to a novel taste and activation peaked at 30 minutes, whereas exposure to a familiar taste had no effect. Conversely, a MAP kinase inhibitor retarded conditioned taste aversion. This combination of findings provides complementary lines of evidence that strongly implicate the NMDA mediated plasticity and subsequent specific protein synthesis as playing a critical role in cortical modifications that mediate this type of learning. Summing up In mammalian systems, the most popular model for the cellular and molecular mechanisms that underlie memory is long-term potentiation (LTP) and its sister phenomenon long-term depression (LTD). Both phenomena follow Hebb's rule in that increases in synaptic efficacy (facilitation of synaptic transmission) marking LTP occur as a consequence of repeated activation of a presynaptic element and its participation in the success in firing the postsynaptic cell, whereas decreases in synaptic efficacy (decrements in synaptic transmission) marking LTD occur as a consequence of the absence of correspondence between activation of a presynaptic element and postsynaptic cell activation. An understanding of the molecular and cellular mechanisms of some forms of LTP is emerging. The induction of one prominent form of LTP involves the activation of NMDA receptors, which occurs when a nonNMDA receptor is initially activated to depolarize the postsynaptic cell, causing the release of a magnesium block of the NMDA receptor allow-

Cellular Mechanisms of Memory: Complex Circuits

77

ing the transmitter glutamate to activate that receptor. This results in an influx of calcium which begins a molecular cascade of events that both stabilizes the changes in the postsynaptic cell and induces gene expression and permanent cellular modifications. Does LTP equal memory, and do we now understand memory in mammalian systems? There is certainly no consensus among researchers that LTP and memory are the same thing, or even that the case for common mechanisms is strong and closed. As described here, there are compelling lines of evidence favoring this view. There is evidence that LTP enhances learning-related information processing, and conversely that learning results in enhancement of synaptic potentials in circuits relevant to particular forms of memory. Correspondingly, there is evidence that blocking LTP with drugs or genetic manipulations can result in a pattern of amnesia reflected both in memory impairments and instability of relevant neural representations. So, while there is also some contradictory evidence, not outlined here, at least a provisional case for shared mechanisms has emerged. In the simpler invertebrate systems discussed in Chapter 2, the circuits that mediated forms of habituation, sensitization, and classical conditioning were mostly identified. So, the additional evidence about cellular changes and their molecular mechanisms seems to offer fairly comprehensive framework for an understanding of memory in those systems. In reading this chapter, though, you were introduced to fragments of a few of the brain circuits involved in learning in mammalian systems, and to a few of the different kinds of learning in which they participate. You have not yet seen the full circuitry involved in any of these systems. Nor have you yet heard about why they mediate different forms of learning, or about how many systems and forms of learning exist. Even for the hippocampus itself, you have only been provided with a glimpse of its role in memory and the nature of the information represented within its circuitry. Completing these stories is a major aim of the remainder of this book. At this point, you should be impressed with the conservation of fundamental cellular and molecular mechanisms of memory across species and brain systems. In all the examples discussed, changes in synaptic efficacy are central. And these changes seem to be subserved by a relatively small set of pervasive molecular mechanisms that reveal a cascade of events that leads from short-term modulation to permanent structural alteration of synapses. Within simpler, well-understood circuitries, this cascade provides a more or less comprehensive picture of memory accomplished. Within more complex circuitries, a higher level of analysis will be required to reach a full understanding of the systems, circuits, and codes for memory.

78

Part I: Connection

READINGS Bear, M.F. 1996. A synaptic basis for memory storage in the cerebral cortex. Proc. Nat. Acad. Sci. U.S.A. 93:13453-13459. Bliss, T.V.P., and Collinridge, G.L. 1993. A synaptic model of memory: Long-term potentiation in the hippocampus. Nature 361: 31-39. Malenka, R.C. 1994. Synaptic plasticity in the hippocampus: LTP and LTD. Cell 78: 535-538. Morris, R.G.M. 1989. Does synaptic plasticity play a role in information storage in the vertebrate brain. In Parallel Distributed Processing: Implications for Psychology and Neurobiology, Morris, R.G.M. (Ed.). Oxford: Clarendon Press, pp. 248-285. Morris, R.G.M., and Frey, U. 1997. Hippocampal synaptic plasticity: Role in spatial learning or the automatic recording of attended experience? Phil. Trans. R. Soc. Lond. 352: 1489-1503. Rioult-Pedotti, M.-S., Friedman, D., Hess, G., and Donoghue, J.P. 1998. Strengthening of horizontal cortical connections following skill learning. Nat. Neurosci. 1:230-234. Rogan, M.T., Staubli, U.V., and LeDoux, J.E.. 1997. Fear conditioning induces associative long-term potentiation in the amygdala. Nature 390:604-607. Silva, A.J., Smith, A.M., and Giese, K.P. 1997. Gene targeting and the biology of learning and memory. Annu. Rev. Gene. 31:527-547. Steele, R.J. and Morris, R.G.M. (1999) Delay dependent impairment in matching-toplace task with chronic and intrahippocampal infusion of the NMDA-antagonist D-AP5. Hippocampus 9:118-136. Stevens, C.F. 1998. A million dollar question: Does LTP = memory. Neuron 20: 1-2.

11 COGNITION: IS THERE A "COGNITIVE" BASIS FOR MEMORY?

W

illiam James may have best captured the essence of the distinction between the behaviorist and cognitivist views of memory in his turn of the twentieth century text, The Principles of Psychology. While acknowledging that habits (learned reflexes) could be chained to mediate even rather complicated and coordinated, sequences, such a form of representation lacked the flexibility and consciousness that characterized real memory. He qualified true memory as "the knowledge of an event, or fact, of which in the meantime we have not been thinking, with the additional consciousness that we have thought or experienced it before" (p. 648). Even in acknowledging that conditioning could revive an image or copy of a prior event, such a revival would not really be a true memory—to be a true memory, the image must not only be conceived of as in the past, but as in one's own past, and the memory must contain one's own experience with the item. This, James argued, would come about only by retrieving memories within a network of associated information, and bringing this network-memory up to the realm of consciousness. But, other than suggesting this would require a vastly complicated brain process, James was at a loss to offer details on the mechanisms of conscious recollection. By contrast, the behaviorist school that emerged in the Golden Era promised tight experimental control and a detailed understanding of the elements of learning within the context of findings emerging from Pavlov's physiological studies. So ignoring James's admonitions, the behaviorists held sway in the thinking of most learning theorists for several decades. Nevertheless, battles between the "behaviorists" and "cognitivists" would continue for that entire period. A major early challenge came from Yerkes, who, based on his extensive observations of problem solving in great apes, 79

80

Part II: Cognition

concluded that higher animals did not learn by random trial and error with reinforcement guiding behavior, but rather exhibited ideation and insight. A greater degree of success in challenging the reductionist approach was achieved by two later experimentalists, Edward Tolman and Fredric Bartlett, working with rats and humans, respectively. Tolman and the "cognitive map" In the 1930s and 1940s Edward Tolman was more successful in challenging behaviorism precisely because he developed operational definitions for mentalistic processes including "purposive behavior" and "expectancy." Moreover, he rigorously tested these ideas using the same species (rats) and maze learning paradigms that were a major focus of prominent behaviorists. Tolman and his students performed several experiments pitting these views against one another in analyses of maze learning by rats. Their studies focused on whether rats could demonstrate "insight" by taking a roundabout route or shortcut in a maze when such strategies were warranted, and were inconsistent with a previous reward history that favored a different route. For example, in an experiment that demonstrated detour taking in rats, Tolman used an elevated maze that involved three diverging and then converging routes from a starting place to a goal box (Fig. II-l). During preliminary training the rat could take any route, and came to prefer the shortest. When this route was blocked (at block A) most rats would prefer to switch to the next shortest route. Only when this route was also blocked would they take the longest path. In the critical test phase, a new block was introduced at the point where the two shorter paths converged (block B). Rats began by running down the shortest path (path 1) as usual. But instead of immediately selecting the next shortest path (path 2), as they had done during the preliminary phase when the shortest route (path 1) was blocked, most rats immediately selected path 3. In addition to this "detour" ability, Tolman also provided evidence that rats could take shortcuts. In this experiment rats were trained to approach a goal via a single circuitous route, then the maze was substituted with many direct paths, some leading toward and others away from the goal locus. Most rats ran to the path that took them directly to where the entrance of the food box had originally been located. These studies provided compelling operational evidence of Tolman's assertions that rats had inferential capacities revealed in the flexibility of the behavioral repertoire that could be brought to bear in solving problems for which behavioral theory had no explanation.

Is There a "Cognitive" Basis for Memory?

81

Figure. II- 1. Schematic diagram of the maze Tolman used to test the ability of rats to infer a detour.

Thus, Tolman's basic premise was that learning generally involved the acquisition of knowledge about the world, and in particular about relationships among stimuli and between stimuli and their consequences, and that this knowledge led to expectancies when the animal was put in testing situations. His views contrasted sharply with those of the behaviorists on three key features of learning. First, the behaviorists argued that the contents of memory involve habits that can be characterized as acquired stimulus-response reflex sequences. But for Tolman learning involved the creation of a "cognitive map" that organized the relations among stimuli and consequences that would guide behavioral solutions to obtain desired consequences. Second, for the behaviorists reinforcement was the central driving force of learning. Thorndike's "law of effect" attributes all learning to the principle that behaviors that lead to a positive reinforcement strengthen stimulus-response connections and are thus more likely to be repeated. Alternatively, Pavlov had proposed that learning involved the association of conditioned and unconditioned stimulus through temporal contiguity. In Pavlov's paradigm, that key association was between the bell, as conditioned stimulus, and the taste of food as unconditioned stimulus; through the learned association the bell comes to substitute for the food taste in

82

Part II: Cognition

evoking salivation. For Tolman, however, reinforcers served simply as more information on which to confirm one's expectancies about when, where, and how rewards were to be obtained—learning itself was driven not by reinforcement but by curiosity about the environment and seeking of knowledge for expectancies about its predictive structure. Reinforcers would certainly determine what behavior might eventually be emitted, but were not necessary for establishment of the representation. Third, for the behaviorists the responses emitted are precisely the motor commands that are the end point of stimulus-response reflexes. The range of behavioral responses then is fully determined as the motor patterns that were elicited and reinforced during learning itself. But to Tolman, learning and performance were fundamentally independent events. That is, what an animal knew about the world and what it was going to do about it were surely related (through its expectancies), but were not the same thing, as the behaviorists held. Thus, Tolman argued, animals could use their cognitive maps and expectancies to guide the expression of learned behavior in a variety of ways not limited to repetition of the behavioral patterns exhibited and reinforced during learning. These aspects of expectancy, insight, and flexibility will prove to be important aspects of modern views of memory. It would be too strong to say that Tolman's arguments were sufficiently compelling to cause learning theorists to abandon the reductionist views. Indeed, as shown by the success of B.F. Skinner's successes in the 1950s and 1960s, the behaviorist school still played out a long run in research laboratories, in the development of some clinical approaches to abnormal behavioral patterns in people, and in the popular press. Frederic Barlett and the "schema" Around the same time as Tolman was carrying out his classic studies on maze learning in rats, the British psychologist Frederic Bartlett published a treatise on human memory. And just as Tolman's theory challenged stimulus-response behaviorism, Barlett's work stood in stark contrast to the then established and better known rigorous methods introduced by Ebbinghaus, which guided the pursuits of most of his contemporary psychologists. However, Bartlett's insights about the structure and richness of memory have proven to be critical for modern views of memory. His work was central in bringing the field of memory research back to issues about the nature of the more complex forms of memory that support conscious recollection. Bartlett differed diametrically from Ebbinghaus in two major ways. First, his interest was in the mental processes used to remember. He

Is There a "Cognitive" Basis for Memory?

83

was not so interested in the probability of recall, as dominated Ebbinghaus's approach, but in what he called "effort after meaning," the mental processing taken to search out and ultimately reconstruct memories. Second, Bartlett shuddered at the notion of using nonsense syllables as learning materials. By avoiding meaningful items, he argued, the resulting memories would necessarily lack the rich background of knowledge into which new information is stored. Barlett's main strategy to study recollection was called the method of repeated reproduction. His best known work examined recollections of a South American Indian folk tale written in a syntax and prose style that were quite different from the culture of his British experimental subjects. In particular the story contents lacked explicit connections between some of the events described, and the tale contained dramatic and supernatural events that would evoke vivid visual imagery on the part of his subjects. These qualities were, of course, exactly the sort of thing Ebbinghaus worked so hard to avoid with his nonsense syllables. But Bartlett focused on these features because he was primarily interested in the content and structure of the memory obtained. Bartlett did not use rigorous operational definitions or statistical measures, but his analyses were compelling nonetheless. Bartlett concluded that remembering was not simply a process of recovering or forgetting items, but rather that memory seemed to evolve over time. Items were not lost or recovered at random, as Ebbinghaus might predict. Rather, material that was more foreign to the subject, or lacked sequence, or was stated in unfamiliar terms, was more likely to be lost or changed substantially in both syntax and meaning, becoming more consistent with the subject's common experiences. These and many other examples led Bartlett to develop an account of remembering known as "schema" theory, in which a schema is an active organization of past experiences in which, during remembering, one constructs or infers the probable constituents of a memory and the order in which they occurred. He proposed that remembering is therefore a reconstructive process and not one of mere reproduction, as Ebbinghaus preferred. In Bartlett's terms, remembering required the ability to "turn round" on one's own schemata, using consciousness to search within the simpler learned sequences for rational and consistent order and to reconstruct them anew consistent with one's whole life of experience. In this way Bartlett gave consciousness a function beyond merely being aware. It played a central role in the reconstructive act of remembering, making it consciously mediated, running contrary to current psychological views that would banish references to consciousness. Ultimately, behaviorist-versus-cognitivist controversy on the nature of memory would be resolved by observations from neuroscience. This work

84

Part II: Cognition

involved examinations of amnesic patients, people who were suffering severe memory loss as a result of specific brain damage. While debilitating, it turns out that the memory capacity lost was selective to aspects of conscious recollection described by James and Bartlett, and had properties shared with Tolman's characterizations of the cognitive map. Other capacities that the behaviorists might recognize as intact stimulus-response learning were spared, even in severe cases of amnesia in humans. Parallel studies on animals with experimental brain damage in the same brain areas implicated in human amnesia provided additional insights into the anatomical psychological bases and fundamental psychological mechanisms of cognitive memory. In addition, related physiological observations provide an understanding of the coding elements that underlie the cognitive mechanisms in conscious memory. These findings are the focus of this section.

READINGS Bartlett, F.C. 1932. Remembering. London: Cambridge University Press. James, W. 1890. The Principles of Psychology. New York: Dover Publications (1950 edition). Tolman, E.G. 1932. Purposive Behavior in Animals and Men. Berkeley: University of California Press (1951 edition).

4 Amnesia— Learning about Memory from Memory Loss

STUDY QUESTIONS Who is H.M. and why is he so valuable to memory research? What nonmemory abilities are spared in amnesia? What memory capacities are spared in amnesia? How is the kind of memory lost in amnesia best characterized?

50 years of experimentation on the course of normal learning did O ver not resolve the debate between behaviorists and cognitivists. But studies on the loss of memory in humans, the phenomenon of amnesia, provided a pair of breakthroughs that has led to an understanding and validation of both the behaviorist and the cognitivist views. The first major breakthrough came with the 1957 report by Scoville and Milner on the most famous neurological patient ever, a man known by his initials H.M. This patient had been severely epileptic for several years. In an effort to alleviate his disorder, the medial temporal lobe area was removed, and indeed the surgery did reduce the frequency of his seizures considerably. However, following the surgery this patient became severely amnesic, and yet showed hardly any other neurological deficits. Because of both the severity and the selectivity of his memory deficit, the findings on H.M. changed everything about how we think about the brain and memory. Before H.M. the search for memories was focused on the 85

86

Part II: Cognition

cerebral cortex—in H.M. the critical damage was in areas underneath the cortex. Before H.M. the generally held view was that memory and other perceptual and cognitive functions were not anatomically separable—H.M. showed a clear dissociation between fully intact perception and cognition versus severely impaired memory. The observations on the pattern of his impairment directly addressed the nature of cognitive processes in memory; these findings are discussed in this chapter in detail. The discovery of a "pure" memory deficit following selective brain damage also addressed how memory is compartmentalized in the brain, and that topic is discussed in the next section of this book. In addition, a prominent component of H.M.'s amnesia is a temporally graded retrograde memory impairment, like that of Ribot's patients introduced in Chapter 1, and the implications of these findings for the phenomenon of consolidation are discussed in greater detail in Chapter 12. The initial observations did not at first provide clarification about the nature of cognitive processes in memory. H.M.'s loss of everyday memory was "global," that is, it appeared to encompass all kinds of memories, and in this broad scope did not directly reveal anything about the memory processes that underlie the deficit. Yet, even from the outset, exceptions to H.M.'s global amnesia were noted—he was able to learn new motor skills. In addition, another hint of an exception to the otherwise pervasive scope of amnesia came from the observation that prior exposure to picture or words could facilitate later identification of those items from fragmentary information. But these spared capacities at first seemed meager compared to the devastation of his overall memory capacity. A deeper understanding about the exceptions came in 1980 when Neal Cohen and Larry Squire made a second major breakthrough in understanding the nature of the memory processing deficit behind amnesia. They described a complete preservation of the acquisition, retention over several months, and expression of a perceptual skill in amnesic patients. The behavioral paradigm they explored involved an improvement in fluency during reading of mirror-reversed words. Ordinarily one is slow in deciphering a word that is presented "backward," that is, with each of the letters reversed as if seen in a mirror. However, with practice one improves considerably at this general skill, even when none of the particular words are repeated. Cohen and Squire found that this kind of learning was fully normal in a set of amnesic patients. In addition, when normal subjects were presented the same mirrorreversed words a second time, they showed an extra level of facilitation in reading them beyond that explained by the general skill acquisition— that is, they showed memory for the particular mirror-reversed words they

Amnesia—Learning about Memory from Memory Loss

87

had seen before. However, the amnesic patients were markedly impaired both in recognizing the familiar words and in recollecting their training experiences. Cohen and Squire were struck by the dissociation between the ability to acquire the general mental procedure of reading reversed text, an ability that appeared fully normal in the amnesic patients, and the capacity to explicitly remember or consciously recollect those training experiences or their contents, which was markedly impaired in the amnesics. They attributed the observed dissociation of these two kinds of memory performance, together with the earlier reported exceptions to amnesia, to the operation of distinct forms of memory. Cohen and Squire suggested that the medial temporal region was specialized for declarative memory, the capacity to consciously recollect everyday facts and events, and that other brain regions were sufficient to mediate a collection of learning capacities that they called procedural memory, the ability to tune and modify the brain's networks that support skilled performance. Two decades of research have supported this dissociation, and further characterized and distinguished the properties of declarative and procedural memory. The following sections provide a more detailed overview of the patient H.M., in order to provide a closer perspective on the nature of his amnesia. Then the distinction between declarative and procedural memory is explored further, using several examples from the experimental literature on amnesia. The amnesic patient H.M. In 1933, when H.M. was 7 years old, he was knocked down by a bicycle, hit his head, and was unconscious for 5 minutes. Three years after that accident he began to have minor epileptic seizures, followed by his first major seizure while riding in his parents car on his 16th birthday. Because of the epileptic attacks his high school education was erratic, but eventually he graduated in 1947 at age 21 with a "practical" course focus. Subsequently, he worked on an assembly line as a motor winder. However, the seizures became more frequent, on average 10 minor attacks each day and a major one each week, and he eventually could not perform his job. Attempts to control the seizures with large doses of anticonvulsant drugs were unsuccessful, leading to consideration of a brain operation. There was no evidence of localization from electroencephalographic (EEG) studies. Nevertheless, because of the known epileptogenic qualities of the medial temporal lobe areas, an experimental operation was considered justified as an effort to ameliorate his devastating seizure disorder. In 1953,

88

Part II: Cognition

when H.M. was 27, Dr. William Scoville performed the bilateral medial temporal lobe resection. The surgical approach to this area is difficult, because the relevant tissue lies inside the part of the temporal lobe near the midline, almost in the center of the brain. The surgical procedure involved making a hole above the orbits of the eyes and lifting the frontal lobes. From this approach the anterior tip of the temporal lobe could be visualized, and the medial part resected. Suction was used to remove all of the tissue bordering the lateral ventricle, including the anterior two-thirds of the hippocampus, as well as the amygdala and surrounding cortex very selectively (Fig. 4-1).

Figure 4-1. A: Position of the hippocampus, amygdala, and surrounding cortex in the human brain. B: Sections through the human brain showing reconstructions of the area of medial temporal lobe removal in the patient H.M., based on MRI scans. Top: A more anterior section showing the area of removal that involved the amygdala (left side) compared to the intact area (right side). Bottom: A more posterior section showing the area of removal that involved the hippocampus (left side) as compared to the intact area on the right. In H.M. the lesions were bilateral (from Corkin et al., 1997).

Amnesia—Learning about Memory from Memory Loss

89

The operation reduced the frequency of seizures to a point that they are now largely prevented by medication, although minor attacks persist. However, one striking and totally unexpected consequence of the surgery was a major loss of memory capacity. Because of the combination of the unusual purity of the ensuing memory disorder, the static nature of his condition, his cooperative nature, and the skill of the researchers who have protected and worked with him, H.M., is probably the most examined and best known neurological patient ever studied. After recovery from his operation, H.M. returned home and lived with his parents. There he did household chores, watched TV, and solved crossword puzzles. Following his father's death he attended a rehabilitation workshop and became somewhat of a handyman, doing simple and repetitive jobs. Eventually his mother and then another relative could no longer care for him, so he was moved to a nursing home where he still resides, participating in daily social activities of the home, as well as watching TV and solving difficult crossword puzzles. He is characterized as a highly amiable and cooperative individual. He rarely complains about anything, and has to be quizzed to identify minor problems such as headaches. He never spontaneously asks for food or beverage, or to go to bed, but he readily follows directions for all of his daily activities. His temper is generally very placid, although this author recalls one day when H.M. was depressed about "having not done anything with his life." However, when assured that he was indeed a very important person, his mood returned to its normal rather upbeat state, and he told one of his famous stories about once considering a career in neurosurgery. He is aware of his memory disorder, but is not consistently concerned about it. Sometimes when given a rather difficult memory question he reminds the tester, "You know I have a memory problem." The severe magnitude of his memory disorder has continued unabated since the time of the surgery. Some of the most compelling examples of the severity of H.M.'s amnesia come from anecdotes of those who have worked with him. I recall my first encounter with H.M., while transporting him from the nursing home to M.I.T. for a testing period in 1980. On the way to the nursing home, I had stopped at a nearby McDonald's for lunch, and had left a coffee cup on the dashboard of the car. When I retrieved H.M., I sat him comfortably in the back seat and we began the trip to Boston. After just a few minutes H.M. noticed the cup and said, "Hey, I knew a fellow named John McDonald when I was a boy!" He proceeded to tell some of his adventures with the friend, and so I asked a few questions and was impressed with the elaborate memories he had of that childhood period. Eventually the story ended and H.M. turned to watch the scenery passing by. After just a few more minutes, he looked up at the dashboard and remarked,

90

Part II: Cognition

"Hey, I knew a fellow named John McDonald when I was a boy!" and proceeded to relate virtually the identical story. I asked probing questions in an effort to continue the interaction and to determine if the facts of the story would be the same. H.M. never noticed he had just told this elaborate tale, and repeated the story more or less exactly as before. A few minutes later the conversation ended, and he turned to view the scenery again. However, just minutes later, once more H.M. looked up to the dashboard and exclaimed, "Hey, I knew a fellow named John McDonald when I was a boy!" I helped him reproduce, as well as he could, the same conversation yet again, then quickly disposed of the cup under the seat. . . . The selective nature of H.M.'s memory disorder

H.M.'s disorder is highly selective in two important ways. First, his impairment is almost entirely selective to memory, as distinguished from other higher-order perceptual, motor, and cognitive functions. Second, even within his memory functions, the disorder is selective to particular domains of learning and memory capacity. Some of the details concerning these two aspects of his preserved and impaired capacities are discussed next. H.M.'s perceptual, motor, and cognitive functions are intact

The results of extensive testing of sensory functions showed that H.M.'s perceptual capacities are entirely normal. He performs well within the normal range on tests of visual acuity, adaptation, and other commonly tested visual-perceptual functions. He can recognize and name common objects. He has some loss of touch and fine motor coordination revealed in sophisticated tests, but these are not noticed in his generally good performance on tasks that require coordination in his daily environment. H.M.'s intelligence was above average in standard IQ tests just before the operation. After the surgery his IQ actually rose somewhat, perhaps because of the alleviation of his seizures. H.M.'s language capacities are largely intact, although he exhibits slight deficits in the fluency of his speech, and his spelling is poor. He appreciates puns and linguistic ambiguities, and communicates well and freely. His spatial perceptual capacities that do not depend on memory are mixed. For example, he has some difficulty copying a complex line drawing, and cannot use a floor plan to walk a route from one room to another in the M.I.T. testing facility. On the other hand, he does well on other complex spatial perceptual tasks, and can draw and recognize an accurate floor plan of his former house. By contrast, H.M. has almost no capacity for new learning, as measured by a large variety of conventional tests. He was not given standard

Amnesia—Learning about Memory from Memory Loss

91

memory tests prior to the operation. After the surgery his scores on standardized scales indicated a severe memory disorder. In particular, he scores zero on components of the test that assess the persistence of his memory for short stories, lists of words or numbers, pictures, or any of a large range of other materials. H.M's memories acquired in childhood are intact, and his immediate memory capacity is normal

H.M. can remember material learned remotely prior to his operation. His memory for the English language seems fully intact. He also retains many childhood memories. By contrast, all memory for events for some period preceding the operation was lost. In addition, H.M.'s immediate or shortterm memory is intact. He can immediately reproduce a list of numbers as long as that of control subjects—thus the "span" of his short-term memory is normal. However, the memory deficit becomes evident as soon as his immediate memory span is exceeded or after a delay with some distraction. These aspects of H.M.'s spared memory abilities are discussed further in later chapters. "Exceptions" to H.M.'s impairment in new learning

The early studies on H.M. also revealed a few "exceptions" to his otherwise profound defect in lasting memory. One of these, called mirror drawing, involved the acquisition of sensorimotor skill. In this task the subject sits at a table viewing a line drawing and one's hand only through a mirror (Fig. 4-2). The line drawing contains two concentric outlines of a star, and the task is to draw a pencil line within the outlines. Errors are scored each time an outline border is contacted. This test may seem simple, but in fact normal subjects require several trials before they can successfully draw the line without committing crossover errors. H.M. showed strikingly good improvement over several attempts within the initial session, and considerable retention of this skill across sessions, to the extent that he consistently made very few errors on the third test day. This success in learning this sensorimotor skill contrasted with his inability to recall ever having taken the test. In addition, H.M. also showed strikingly good performance in perceptual learning in a task called the Gollins partial pictures task, which involves the recognition of fragmented line drawings of common objects. For each of 20 items, subjects are presented with a series of five cards containing fragments of a realistic line drawing of the same object. The first card of each series contains the fewest fragments of the drawing and the last card contains the complete drawing (Fig. 4-3). Subjects are initially

92

Part II: Cognition

Figure 4-2. The mirror drawing test. Top: Sketch of the double-star pattern and the beginning of a typical early attempt at drawing a line between the boundaries. Bottom: H.M.'s performance across 10 attempts on 3 successive days (data from Milner et al., 1968).

shown all 20 of the most difficult items and asked to identify the object drawn on each one. Then the second, slightly more complete version of each item is presented with the ordering of the 20 cards randomized, so it is impossible to anticipate an item based on its predecessor. The procedure is continued using successively more complete versions of each item until all are identified. Then, after an hour of intervening activity, the entire test is repeated, and the number of errors (unidentified drawings) is scored. Normal subjects show retention of this perceptual memory reflected in the ability to identify less complete versions of the drawings. H.M.'s scores on the retest were not as good as age-matched controls, but he showed a surprising degree of retention, especially considering he did not remember having taken the initial test. A broad range of spared learning abilities in amnesia The observations of Cohen and Squire showing fully intact perceptual skill learning were properly heralded as a revelation about memory processing

Amnesia—Learning about Memory from Memory Loss

93

figure 4-3. The Gollins partial pictures test. A: An example of the series of partial figures that compose one item on the test. B: Performance of H.M. and normal control subjects measured by the number of errors on the entire series on the initial test and on a delayed retest with the same items (data from Warrington and Weiskrantz, 1968).

functions accomplished by the medial temporal lobe. Subsequently, several laboratories uncovered a variety of examples of spared learning ability in amnesia. Examples from a broad range of those findings are presented next, to provide a view of the scope of preserved learning and memory capacities observed in amnesia. These include a form of perceptual learning called "priming," skill learning, Pavlovian conditioning, and sequence learning. Priming Perhaps the most intensively studied form of memory that can be accomplished fully normally in amnesic patients is the phenomenon known as repetition priming, or just "priming." Priming involves initial presentation of a list of words, pictures of objects, or nonverbal materials, and then subsequent reexposure to fragments or very brief presentation of the whole item. In the reexposure phase, learning is measured by increased ability to reproduce the whole item from a fragment (as in the Gollins partial pictures task described earlier) or by increased speed in making a decision about the item.

94

Part II: Cognition

One example study particularly nicely illustrates a striking dissociation between intact priming and impaired declarative memory performance by amnesic subjects. This experiment used the word stem completion task, a test of verbal repetition priming in which subjects initially study a list of words, then are presented with the first three letters of each word (the word "stem") and asked to complete it. The stimulus words are selected as ones for which the stem can be completed more than one way to compose a high frequency word. For example, the word "MOTEL" is used because its stem "MOT " can be completed to form either the stimulus word or "MOTHER" (see other examples in Fig. 4-4 top). Priming is measured by the increased likelihood that the subject will complete the stimulus word presented during the study phase. In this experiment, subjects initially studied a list of such words and, to make sure they attended to them, had to identify shared vowels among sets of words or rate the words according to how much they liked them. Then, in the test phase, they were presented with the three-letter word stems and tested for their memory in one of three ways. In the free recall condition, subjects were not presented with stems but just asked to recall the studied words. In the cued recall condition, subjects were presented with the word stems and told to use them as cues to remember words that were on the list. In the completion condition, they were presented with word stems and asked sim-

figure 4-4. Word stem completion test of verbal priming. Top: On the left are examples of study words, and on the right, examples of the word stems used for cueing. Bottom: Performance of normal control subjects and amnesics on three versions of the test (data from Squire, 1987).

Amnesia—Learning about Memory from Memory Loss

95

ply to "write the first word that comes to mind." The amnesics were impaired in recall as tested either with cueing or without (Fig. 4-4 bottom). By contrast, they were not impaired on the completion test. A particularly revealing comparison can be made between the performance of amnesics across the different test conditions. They did much better on the cued recall than the free recall condition, but no better on cued recall than on completion. One interpretation of these findings is that performance in cued recall might be entirely supported by priming. By contrast, the normal subjects did much better on cued recall than on priming, suggesting they used the stems to aid an active search in recalling the words. Intact priming in amnesia is not restricted to nameable objects and verbal material. For example, H.M. also shows normal priming in a task explicitly designed to be refractory to verbalization. In this test H.M. and normal subjects were presented with a set of stimuli each of which consisted of five dots arranged in a unique pattern (Fig. 4-5 top). To establish baseline performance, subjects were asked to draw on the dots any line pattern they wished. Substantially later they were presented with a set of predetermined target patterns and asked to replicate them onto a corresponding dot pattern. After exposure to the entire list plus a distracter task, they were provided with the dot pattern again and asked to complete it any way they wished. Priming scores were calculated based on the incidence of baseline patterns. As shown in Fig. 4-5 (bottom), H.M. showed significant above chance priming for dot patterns, indicating as much memory as the normal subjects. This intact performance stood in contrast to his inability to recognize the same dot patterns when explicitly asked if he had seen them before. Skill learning Mirror drawing, described before, is an example of spared capacity for the acquisition of sensorimotor skills. In addition, the intact capacity to learn skills extends to the acquisition of cognitive rules. For example, in one study subjects were presented with strings of letters that were generated by an artificial "grammar" that determined general rules for sequencing and length of the letter strings (Fig. 4-6). They studied these strings by reproducing each item immediately after its presentation. Then subjects were informed that the letter strings were formed by complex rules. Subsequently they were shown novel letter strings one at a time and asked to classify them as "grammatical" or "nongrammatical" according to whether they conformed with the rules. Finally, subjects were tested to determine if they could recognize grammatic letter strings after a brief study phase. Both amnesic and normal subjects were able to correctly classify the letter strings

96

Part II: Cognition

Figure 4-5. Priming for dot patterns test. Top: Examples of dot patterns and target completion pattern. Bottom: Performance (percentage correct) as measured by recognition of the target and by correct completion in normal control subjects and case H.M. (data from Gabrieli et al., 1990).

on about two-thirds of the trials. By contrast, the amnesic patients were impaired on recognition of studied grammatical items. Classical (Pavlovian) conditioning

Modern formal studies of classical conditioning in both humans and animals have focused on conditioning of eyeblink reflexes, because these are easy paradigms to control and allow straightforward measures of learning. These studies typically involve repeated pairings of a tone or light as the conditioning stimulus (CS) and an airpuff to the eye as the unconditioned stimulus (US) that produces a reflexive blink. The measure of classical conditioning is the occurrence of eyeblinks during the CS period prior to presentation of the US, that is, conditioned eyeblink responses. Systematic studies comparing amnesic and normal subjects have demonstrated intact classical eyelid conditioning in amnesics, as well as normal extinction of conditioned responses when the CS was presented repeatedly with-

Amnesia—Learning about Memory from Memory Loss

97

Figure 4-6. Artificial grammar test. A rule system used to string letters into "grammatical" and "nongrammatical" series, and several example items (data from Knowlton et al., 1992),

out the US. The examples of intact eyelid conditioning all involve a procedure known as "delay-conditioning," in which the presentation of the CS is prolonged and overlaps the US. Another procedure, known as "traceconditioning," involves a brief CS followed by a trace interval during which no stimulus is presented, followed by the US alone. The distinction between these two types of classical conditioning is important because it has been shown that rabbits with hippocampal damage normally acquire the eyelid response with the delay-conditioning procedure but cannot learn under the trace-conditioning procedure. Just like rabbits with hippocampal damage, human amnesics are impaired in trace eyelid conditioning. Moreover, this deficit has been related to the conscious awareness of the stimulus contingencies. Sequence learning Another domain of intact learning in amnesia involves the gradual improvement in speed of performance following specific regularities in stimulus-response sequences. A compelling example of intact learning of a specific perceptual-motor habit involves manual sequence learning in a task called the serial reaction time test. On each trial subjects were shown a light at one of four locations on a computer monitor, and had to press one of four keys that corresponded to that light location. On each trial the lights were presented in a consistent pattern, and the entire pattern was presented repetitively in each training session such that subjects could anticipate the position of the next light. As a control, in separate testing blocks the sequence of light positions was randomized. Normal subjects and amnesics decreased their reaction times to press the keys, and did so at an equal rate. Both groups showed minimal improvement when the light

98

Part II: Cognition

position sequence was random, indicating that the improvement on regular sequences involved acquisition of a specific habit and not a general ability to coordinate the pressing of keys for appropriate lights. Another example of spared learning of a specific habit in human amnesic patients comes from studies of speed reading. Several experiments have shown that amnesics improved their reading times over the course of repeating a story aloud. In these studies there was no general facilitation between stories, indicating that the habit was text-specific. Moreover, this intact capacity does not seem to rely on memory for the content of the story, as demonstrated in an experiment showing that the phenomenon extends to nonwords (pronounceable but meaningless letter strings) as well as text. However, even with nonwords, the facilitation is specific to the sequence of repeated nonwords and is not a reflection of a general learning to read new nonwords. Characterizing the properties of declarative and procedural memory There have been numerous attempts to identify the common properties among the types of learning and memory spared in amnesia, and to distinguish them from the common aspects of learning and memory on which amnesics fail. These comparisons have provided insights into the nature and cognitive mechanisms that underlie declarative memory, as well as properties of the domains of procedural memory. One way to characterize declarative memory is to consider whether there is a set of common or fundamental properties of memory that is spared in amnesia. Some of the earliest findings on H.M. indicated that intact learning in amnesia is limited to motor skills or simple perceptual learning. Other studies have suggested that spared learning always involves the slow incremental acquisition of habitual routines or to specific categories of information. However, intact learning in amnesia is not limited to general motor skills. Rather, the scope of spared memory in amnesia includes a variety of forms of highly specific new learning. Also, intact learning in amnesia is not limited to simple forms of perceptual learning or other easy tasks, as clearly shown by the improvements in very difficult tasks such as grammar learning and sequence production. In addition, spared learning in amnesia is not limited to types of learning that involve slow incremental improvement, but includes a variety of forms of one-trial learning, such as observed in numerous repetition priming tests. Indeed, priming for single exposures to pictures can be both robust and last at least a week in am-

Amnesia—Learning about Memory from Memory Loss

99

nesic subjects. Finally, preserved learning in amnesia is not limited to any particular category of learning materials, as one can see in our broad range of examples, including intact learning for words, nonwords, common objects, tones, nonverbalizable pictures, motor patterns, and spatial patterns. In sum, the domain of intact learning ability is global, and it can be either fast or slow, and include both general skills and highly specific information content. Thus, the common properties among examples of intact memory in amnesia are not to be found in relatively objective parameters such as the modality of information, or the speed or specificity of memory. Formal characterizations of nondeclarative memory Several hypotheses have emerged from efforts to characterize the critical features that are common to intact learning in amnesia. Instead of simpler objective parameters such as those listed just previously, these proposals focus on more complex and higher-order properties, and in particular, the form of memory expression, the extent of conscious access to memories, and the structure of the memory representation. One of the most objective attempts to formalize the general learning abilities of amnesics was provided by Morris Moscovitch in his 1984 list of sufficient conditions for demonstrating preserved memory. He focused on task demands, and argued that amnesics show savings on tests that satisfy three conditions: (1) the task has to be so highly structured that the goal of the task and the means to achieve it are apparent, (2) the means to achieve the goal are available to the subject (i.e., the response strategies are already in the subject's repertoire), and (3) success can be achieved without reference to any particular event or episode. Combining these, he suggested that amnesics will succeed whenever they simply have to perform a task guided by the conditions and strategies at hand. The new memory is revealed in changes of task performance itself, typically either a change in the speed of responding or in a bias of choices that are readily available. Daniel Schacter made a similar distinction between "explicit" and "implicit" memory. Explicit memory involves conscious recollection generated by direct efforts to access memories. Explicit tests of memory involve direct inquiries that ask the subjects to refer to a specific event of learning or a specific fact in their knowledge. Examples of explicit tests of memory include, "What were the words on the list you studied?" and "Which of these two items did you see before?" The full range of explicit memory tests includes a large variety of direct measures of recall or recogni-

100

Part II: Cognition

tion of word or picture lists, paired associates, story recall, and most of the common tests of memory that are performed so poorly by amnesic patients. Explicit memory expression also includes most everyday instances of memory, such as recalling what one had for breakfast this morning or what the capital of France is. Both examples involve conscious efforts to search for a specific event or fact. By contrast, implicit memory involves unconscious changes in performance of a task as influenced by some previous experience. Implicit tests of memory involve indirect measures such as changes in the speed of performance or in biases in choices made during performance of a task that can be solved with the information at hand. Examples of implicit memory tests include the full variety of assessments of motor, perceptual, and cognitive skills, habits, conditioning, and repetition priming described earlier at which amnesic patients usually succeed. Notably none of these tests requires the subjects to be aware of their memory, or to "remember," a specific event or fact. A related proposal is Endel Tulving's distinction between "episodic memory" and "semantic memory." Episodic memory contains representations of specific personal experiences that occur in a unique spatial and temporal context. Episodic memory involves the capacity to reexperience particular events in one's life, what Tulving calls "time traveling." By contrast, according to Tulving, semantic memory is the body of one's world knowledge, a vast organization of memories not bound to any specific experience in which they were acquired. Some investigators have suggested that the pattern of impaired and spared memory capacities in amnesia can be explained as an impaired episodic memory capacity and intact semantic memory. This view readily accounts for the impairment in day-to-day episodic memory ("What did you have for breakfast?"). In addition, episodic memory can strongly facilitate one's performance on many standard tests, such as one's ability to recall or recognize a recently studied list of words. Memory in these situations is stronger in normal subjects because they can refer to their episodic memory for the specific learning experience, in addition to any memory for the materials independent of that specific experience. But, according to this view, amnesics have only the episode-free record and so usually perform less well. Furthermore, according to this view, amnesic subjects perform well on implicit memory tasks, because the memory demands avoid reference to the temporal or spatial context in which the information was acquired, and do not require the subject to refer to the learning experience directly. In these tests there is typically no advantage conferred on normal subjects in remembering the items. The episodic-semantic distinction shares much

Amnesia—Learning about Memory from Memory Loss

101

with Moscovitch's characterization of successful memory performance whenever amnesics do not have to "conjure up, that is, 'remember,' any previous experience or a newly learned fact." Indeed, the episodic-semantic and implicit-explicit views are fully compatible, to the extent that implicit memory tests always and only require semantic memory. In support of this view, Tulving et al. described a patient, K.C., with normal intelligence, preserved general knowledge, and fragmentary general knowledge of his past. He also had expert knowledge from work done 3 years before a closed head injury. By contrast, K.C. did not remember a single personal event from his previous life and did not remember new events. He had some capacity to gradually acquire new knowledge, as demonstrated in studies aimed at very gradual accumulation of semantic knowledge by teaching methods that reduce interference associated with making errors. In addition, there are now several cases of childhood brain injury that result in amnesia for everyday life events, but near normal general world knowledge. Some of the latter cases appear to have relatively circumscribed damage to the hippocampus, suggesting specific involvement of this structure in episodic memory. Disentangling episodic and semantic memory is a difficult problem. Surely these patients forget "facts," such as a list of words, just as rapidly as they forget daily events, such as what they had for breakfast. One study directly addressed the issue of semantic learning in amnesia by attempting to train H.M. on new vocabulary words. H.M. and normal subjects were given implicit test instructions and were directed away from conscious recollection of the events surrounding the learning experiences. Training proceeded in several phases. They first studied word definitions for eight novel vocabulary words created by the experimenters. The subjects were given a recognition test asking them to choose a definition for each word. Then they studied synonyms for the words, and were tested on a sentence completion task where they had to fill in a blank at the end of a sentence with one of the new words. Normal subjects learned the new words readily, completing each phase within a few trials. Despite an exhaustive regimen of testing, H.M. showed virtually no ability to learn new semantic knowledge. This is not to say that amnesics cannot acquire any semantic knowledge, and indeed there are many examples of highly specific learning even for complex materials. For example, there have been demonstrations of successful learning where subjects were trained to use computer commands and terminology. The training methods used painstaking and very gradual procedures by which the commands were introduced in situations where responses were "error-free" at each stage. These subjects did sub-

102

Part II: Cognition

sequently show aptitude for learning new computer terms. But the range with which they could use this new learning was highly limited, such that their learning could be expressed only in replications of the precise training conditions. Schacter referred to this characteristic of their successful learning as a "hyperspecificity" of the preserved memory. Such limited applicability is not characteristic of our common use of semantic memory in solving everyday problems across a broad range of one's daily challenges. Summing up H.M. was important as much for the selectivity of his deficit as for its severity. Subsequent to the discovery of H.M.'s amnesia, Scoville publicized the findings, and supported Milner, Corkin, and others in their research on H.M., in great part to insure that the operation would not be performed again. H.M. was among a group of patients who underwent the experimental operation for bilateral medial temporal lobe resection. However, all the other patients were severely psychotic, muddling the interpretation of the memory tests. The resection in some of the patients involved only some of the cortex and the amygdala, and these patients' memory was intact. Also, the severity of the amnesic deficit in other patients was related to the amount of hippocampal damage, so it was concluded that the hippocampus and immediately adjacent cortex were the likely critical area for memory. Combining the data across an enormous range of memory and nonmemory assessments, H.M.'s amnesia is characterized by several cardinal features: (1) intact perceptual, motor, and cognitive functions, (2) intact immediate memory, (3) severe and global anterograde amnesia, (4) temporally graded retrograde amnesia, (5) spared remote memory. At that time, views about memory were most influenced by the notion that different cortical areas contained specific perceptual and memory functions together, such that perception, cognition, and memory were considered inseparable and, by Lashley's proposal, widely distributed in the brain. The case study of H.M. was a breakthrough because it showed that a general memory function could be dissociated from other functions. In addition, the findings of exceptions to severe global amnesia, in successful sensorimotor and perceptual learning, foreshadowed a second major breakthrough that promises to further clarify the nature of hippocampal processing in memory. Other studies on many amnesic patients have shown that the domain of spared learning in amnesia includes intact repetition priming, skill learning, Pavlovian conditioning, sequence learning, and more. The common

Amnesia—Learning about Memory from Memory Loss

103

features that distinguish the impaired and preserved memory capacities in amnesia have been characterized in several ways, including the distinctions between explicit and implicit memory expression, and between episodic and semantic memory. Based on considerations from a wealth of data from studies on amnesic patients, the most consistent characterization of the domain of memory impaired in amnesia is captured by the notion of "declarative memory," the memory for facts and events that can be brought to conscious recollection and can be expressed explicitly. Conversely, the most consistent characterization of learning ability spared in amnesia is the notion of "procedural memory," the acquisition of skills and preferences that can be expressed unconsciously by implicit changes in the speed or biasing of performance during a repetition of processing of the learning materials. These characterizations capture all of the features of the distinctions outlined previously. However, they leave unresolved the nature of memory traces that underlie either category of memory. An understanding memory representation at that more fundamental level requires the establishment and exploitation of animal models of different types of memory, because only in such models can the required biological recordings and manipulations be pursued. A consideration of the challenges and successes of animal models of declarative and procedural memory begins in the next chapter.

READINGS Cohen, N.J. 1984. Preserved learning capacity in amnesia: Evidence for multiple memory systems. In The Neuropsychology of Memory, N. Butters, and L.R. Squire, (Eds.) New York: Guilford Press, pp. 83-103. Cohen, N.J., and Squire, L.R. 1980. Preserved learning and retention of a pattern- analyzing skill in amnesia: Dissociation of knowing how and knowing that. Science 210:207-210. Corkin, S. 1984. Lasting consequences of bilateral medial temporal lobectomy: Clinical course and experimental findings in H.M. Semin. Neurol. 4:249-259. Moscovitch, M. 1984. The sufficient conditions for demonstrating preserved memory in amnesia: A task analysis. In The Neuropsychology of Memory, N. Butters, and L.R. Squire (Eds.). New York: Guilford Press, pp. 104-114. Ogden, J.A., and Corkin, S. 1991. Memories of H.M. In Memory Mechanisms: A Tribute to G.V. Goddard, W.C. Abraham, M. Corballis, and K.G. White (Eds.) pp. 195-215. Schacter, D.L. 1987. Implicit memory: History and current status. /. Exp. PsychoL Learn. Mem. Cogn. 13:501-518.

104

Part II: Cognition

Scoville, W.B., and Milner, B. 1957. Loss of recent memory after bilateral hippocampal lesions. /. NeuroL Neurosurg. Psychiatry 20:11-12. Squire, L.R., Knowlton, B., and Musen, G. 1993. The structure and organization of memory. Annu. Rev. Psychol. 44:453-495. Tulving, E. Schacter, D.L., McLachlin, D.R., and Moscovitch, M. 1988. Priming of semantic autobiographcal knowledge: A case study of retrograde amnesia. Brain and Cogn. 8:3-20.

s Exploring Declarative Memory Using Animal Models

STUDY QUESTIONS What is an animal model of amnesia? Why are such models valuable? What characteristics of amnesia are well modeled using nonhuman primates? What characteristics of amnesia are well modeled using rodents? How good were the initial attempts at each of these models? What advances led to breakthroughs in each model?

A

lmost immediately after the early reports on H.M. and other patients suffering the consequences of medial temporal lobe excision, efforts began to reproduce elements of the amnesic syndrome in monkeys, rats, and other animals. The major aim of these early efforts was twofold. First, specific experimental brain damage offered an increase in anatomical specificity over that which occurs in cases of surgeries, accidents, and disease. Increased anatomical specificity of the damage improved the ability of investigators to designate which structures of the temporal lobe are critical to memory. Second, in animals, to a much greater extent than in humans, investigators can control the nature and extent of experience gained prior to the brain damage. Human patients arrive in the clinic with a unique background of learning that differs along many dimensions and to a very great extent among individuals. In addition, the specifics of the history of 105

106

Part II: Cognition

individual people can only be known to the extent that there is a record of one's history, which is typically rather vague. By contrast, in animals, investigators can dictate all the details of experience the animal brings to the experimental setting where learning will be studied. From many of our considerations so far, it should be obvious that new learning occurs in the context of previously acquired knowledge (one's schema or prior semantic knowledge), and so it is greatly advantageous to know and control the nature and extent of that knowledge. As in any other situation where an experimental model is desired, it is important at the outset to classify the precise aspects of the human condition that one wishes to model. Fortunately, the properties of a valid animal model of the human amnesic syndrome were clearly outlined in the clinical studies presented earlier: (1) sensory, motor, motivational, and cognitive processes should be intact; (2) short-term memory should be intact. (3) following preserved short-term performance, memory should decline with abnormal rapidity, that is, exceeding the rate of natural forgetting in intact control subjects. (4) the deficit should be global in scope for the tobe-learned materials, that is, the impairment should span sensory and conceptual modalities of new learning. (5) there should be a graded retrograde impairment, such that learning accomplished recently prior to brain damage would be lost, whereas learning accomplished remotely long before the damage should be spared.

Two lines of research in the development of animal models The efforts to model amnesia associated with damage to the medial temporal lobe followed two parallel approaches, one using monkeys as the experimental subjects and the other using primarily rats. The studies on monkeys began appropriately by reproducing the same pervasive medial temporal damage that was done to H.M. Therefore, this line of research has been most useful in characterizing the nature of the memory mediated by the entire set of structures in the medial temporal lobe. The early studies on rats focused on the hippocampus, leaving out of the experimental ablation other structures that were damaged in H.M. and in experiments on monkeys. Therefore, this line of research has been most useful in characterizing the role of the hippocampus itself. The conclusions derived from these two lines do not entirely overlap, because of differences both in the size and locus of experimental brain damage and in the behavioral tests typically employed in monkeys and rats. The following sections summarize some of the findings from both lines of research.

Exploring Declarative Memory Using Animal Models

107

The development of a model of amnesia in monkeys The behavioral assays initially used to study the role of the medial temporal lobe in monkeys focused on two type of tasks that were already being used in comparative studies on the cognitive functions of nonhuman primates: visual discrimination and matching to sample. In visual discrimination training, typically there are two stimuli for each problem, usually flat plaques painted with different colors or patterns, or easily discriminated three-dimensional objects (Fig. 5-1A), each placed to cover food wells on a choice platform. One stimulus is arbitrarily assigned as "positive," and displacing the plaque would reveal a hidden reward on each trial. The other stimulus is assigned a "negative" value and never rewarded. The positions of the stimuli are varied randomly across trials, so their spatial position does not correlate with the reward locus. In delayed matching to sample, each trial is composed of three distinct phases (Fig. 5-1B). In the first phase, called the sample phase, a single stimulus is presented. This is followed by a variable delay phase during which the monkey has to remember the stimulus for different periods of time. In the third phase,

Figure 5-1. Illustration of trials in four memory tests used on monkeys: A: Visual discrimination. B: Delayed match to sample with trial-repeated stimuli. C: Delayed nonmatch to sample test with trial-unique objects as memory cues. D: Delayed spatial response. ITI = intertrial interval.

108

Part II: Cognition

called the choice phase, two stimuli are presented, one identical to the sample and baited and the other different and not baited. So the requirement of the task is to "match" a choice stimulus to the sample in order to receive the reward. Typically in these early studies, the same two stimuli were two-dimensional patterns reused on every trial, with the sample selected randomly across trials. As described later, there are several important variations on this task, employing different kinds of stimuli, a different "nonmatching" rule by which the subject must select the alternative to the sample item in the choice phase (Fig. 5-1C). Among the variations in stimuli was the use of two identical plain stimuli, and the requirement was for the subject to select the same location it had seen food covered by the plain stimulus on the sample trial. This is called the classic "delayed spatial response" task (Fig. 5-1D). The early efforts to model amnesia in monkeys using these tests were not impressively successful. The general pattern of results was not inconsistent with the properties of amnesia listed earlier. Also, some of those key properties, including no effect on aspects of the task that did not require memory and normal short-term memory, were very well duplicated in monkeys with medial temporal lobe damage. But the magnitude of both the anterograde and retrograde components of the memory deficit were quite modest compared to the apparent almost total loss of memory observed in H.M. Monkeys with substantial removals of all of the medial temporal lobe structures were only mildly impaired on learning new visual pattern, color, object, or auditory discrimination problems. In relearning visual pattern or object discriminations that the monkeys had acquired a few weeks prior to the surgery, deficits were reliably observed. However, the magnitude of this retrograde impairment was also disappointing—monkeys with medial temporal lobe damage merely showed less savings from the previous learning and not a complete loss of recently acquired information. Furthermore, monkeys with medial temporal lobe lesions performed surprisingly well on matching to sample and other delayed response tests. The task was trained preoperatively, and there was a retrograde impairment in reacquisition of the task with short delays. This loss of recent memory was consistent with the characteristics of human amnesia. However, having reacquired the task after the surgery, the monkeys performed well even at memory delay intervals of several seconds. It was a state of affairs that led many to suggest that there might be a true species difference in the role of the medial temporal lobe in memory, such that memory relied much more on hippocampal function in humans than in animals.

Exploring Declarative Memory Using Animal Models

109

Success with a new test of recognition memory

A major breakthrough came with a combination of a novel twist in the procedures used for the delayed matching to sample task combined with a modification in the approach for removing medial temporal lobe structures. The key aspects of the task variant involved the use of new sets of stimulus objects on each trial (the "trial-unique stimulus" procedure) plus a nonmatching reward contingency (Fig. 5-1C). Thus, on each trial, the sample was a novel three-dimensional "junk" object. Then, to obtain a reward during the choice phase, the subject was required to select a different novel junk object over the now-familiar sample object. Because an entirely novel sample is used on each trial, it is appropriate to think of this task as a test of recognition for the newly familiar object. Notably this characterization of the task is quite different from that for the task where the same stimuli are used repeatedly—in such a situation both stimuli are highly familiar, so their potential for recognition would hardly differentiate them. The distinction between the trial-unique or repetitive stimulus procedures had a profound effect on monkeys' performance in the delayed matching (or nonmatching) task and on the effects of medial temporal lobe damage. Normal monkeys learned the task with the trial-unique procedure exceedingly rapidly. Monkeys with damage to the hippocampal region were impaired in learning the task when the memory delay was short, but they did eventually reach a high performance criterion and continued to perform well when the memory load was low. However, when the delay was extended, a deficit was observed and the severity of the impairment increased as the delay was elongated (Fig. 5-2A). In addition, if a list of items was presented and then memory for each was tested in a sequence of choice trials, a severe deficit was observed. Monkeys with damage to the medial temporal lobe have a selective memory impairment

The introduction of a new benchmark assessment of amnesia monkeys using the delayed nonmatch to sample (DNMS) task opened up the opportunity to readdress whether this approach would indeed provide a valid model of the fundamental characteristics of human amnesia. Recall that these characteristics include: spared nonmemory functions and short-term memory in the face of rapid forgetting, global scope of amnesia across learning materials, and graded retrograde amnesia. A central issue is the selectivity of the deficit to memory and the sparing of perceptual, motor, motivational, and attention or other cognitive functions. The DNMS task provides an automatic and ideal control in that

110

Part II: Cognition

Figure 5-2. Performance of normal monkeys and monkeys with medial temporal damage on different memory tasks. N = normal animals; H-A = animals with hippocampus plus amygdala and cortical damage (data from Squire, 1987).

all of those nonmemory functions are fully required even in the absence of a memory delay. Thus, if monkeys with medial temporal lobe damage perform normally at the shortest delay, it must be that they can attend to, perceive, and encode the object cues, that they can execute the choice responses, that they are motivated to participate in the task, and that they can acquire and retain the nonmatching rule. Producing a DNMS task with no memory delay at all is problematic because, quite simply, the manually operated apparatus typically used requires 8-10 seconds to exchange sample and choice objects, and deficits in memory can be apparent in human amnesics within 10 seconds. This issue was addressed directly by the development of a computerized version of the DNMS task that employed complex visual patterns presented on a "touch screen" of a video display. Intact subjects required trials to learn the computerized version of the task at the brief delay. Nevertheless, following a medial temporal lobe ablation the rate of learning was fully

Exploring Declarative Memory Using Animal Models

111

normal. In subsequent testing with delays up to 10 minutes, normal monkeys showed a gradual forgetting. Monkeys with medial temporal damage performed as well as normal subjects at delays up to 1 second, but at longer delays a deficit became ever more apparent. Another characteristic of human amnesia is that the deficit can be seen across a variety of learning materials. That is, the impairment is global with regard to the stimulus modality of items to be remembered. Early evidence that addressed this question came from studies of the delayed spatial response task, which is based on memory for a spatial location and not the visual qualities of a particular stimulus. Monkeys with medial temporal damage show the same spared memory at brief delays and increasing impairment at long delays as seen in standard DNMS (Fig. 5-2B). In addition, this issue was addressed by the development of a tactual variant of the DNMS task. Monkeys were initially trained on the conventional version of the task, and then retrained with the room lights gradually dimmed to complete darkness except for small dim cue-lights signaling the positions of the objects. In this situation the animals had to perceive and encode the objects entirely by tactual cues. Normal monkeys required about twice as many trials to relearn the task in the dark, but performance was excellent in the final preoperative stage of tactual DNMS testing even over long delays. After removal of the medial temporal lobe, relearning at a short delay was substantially impaired, but the animals did eventually succeed and continued to perform well with short delays, as observed with the visual version of the task. More important, the deficit grew as the delay was elongated. Thus, the pattern of sparing of ultimate performance at a short delay and increasing impairment at longer delays was identical to that observed for the visually guided version of the same task, demonstrating that the amnesic deficit extended across specific sensory modalities. Another central characteristic of the amnesic syndrome is the phenomenon of graded retrograde memory loss. As described in Chapter 4, H.M. and other amnesic patients display a loss of memories backward in time from the moment of brain damage, with the most severe loss in the period just prior to the damage and total sparing of remotely acquired memories, including childhood recollections and general knowledge acquired early in life. However, a major problem in the interpretation of retrograde memory loss in human patients is that the amount and timing of prior learning experiences can only be approximated. Conversely, one of the major advantages of an animal model of amnesia is that one can examine the retrograde loss of memories with a "prospective" design. That is, one can provide measured amounts of learning at specific times prior

112

Part II: Cognition

to brain damage that occurs suddenly. In so doing, one can directly compare the strength of memories acquired at different times in intact animals and one can more accurately measure the period and magnitude of retrograde loss. Unfortunately, the DNMS task is unsuitable for this kind of study because single exposures to objects do not provide sufficiently strong memories to endure testing weeks or months after learning. To address this issue and develop a task that would be suitable for a study of retrograde memory, experimenters created an object discrimination task where subjects were presented with pairs of novel objects like those used in DNMS testing, and repeatedly reinforced the choice of one object over the other. Normal monkeys learned sets of these pairings rapidly, accumulating 100 successfully acquired problems over 4 months of training. Following completion of the learning series, in half of the animals the hippocampus and nearby cortex were removed and the animals were allowed 2 weeks to recover. Then all the animals were each tested with just one trial on each problem. Normal animals scored well on the most recently acquired problems, and their performance declined a bit, showing some forgetting for problems learned more than 2 months before. By contrast, monkeys with hippocampal damage were substantially impaired, performing at just above that expected by chance, on problems presented within 2 weeks of the surgery. They performed significantly better on remotely learned discriminations, exhibiting normal performance on those acquired 4 months prior to the surgery. This pattern of recent retrograde memory loss and spared remote memory, emphasized most strikingly by worse performance on recent memory than remote memory within the medial temporal group, provides compelling evidence that damage to the medial temporal region results in a graded retrograde amnesia (see Chapter 12 for an extended discussion of this topic). A domain of spared learning ability in monkeys with medial temporal damage

In addition to the previously described aspects of impaired learning and memory following hippocampal damage, there is the critical characterization of a spared domain of new learning capacity in H.M. and other amnesic patients. Toward the goal of modeling this phenomenon, there are specific examples that represent a domain of spared learning in monkeys following ablation of the entire medial temporal lobe. One spared domain that closely parallels intact motor skill learning in human amnesics is the acquisition of manual skills in monkeys. In a study of this kind of learning, experimenters devised two manual skill tests on

Exploring Declarative Memory Using Animal Models

113

which monkeys could be trained. One of these involved training the monkeys to reach around a clear barrier to obtain a reward. Another involved challenging the monkeys to obtain a doughnut-shaped candy (a "lifesaver") reward that was presented in the middle of an irregularly bent stiff wire (a coat hanger). To rapidly retrieve the reward the monkey had to improve its manual manipulation of the lifesaver around the turns of the coat hanger. In both tasks monkeys with medial temporal damage improved in performance at the same rate as normal subjects, demonstrating preserved motor skill learning in amnesia (Fig. 5-2C, D). In addition, monkeys with medial temporal lobe damage perform normally well in the acquisition and retention of single visual discrimination problems that are acquired gradually. This spared learning capacity was described in some of the early studies on medial temporal lobe ablations in monkeys, and was confirmed in studies using the same junk object stimuli employed in DNMS. Notably, the deficit is observed only under conditions where normal learning was slow and gradual (by presenting each pair of object only once per day). In conditions where normal animals learn object discriminations most rapidly (acquisition in a single session with multiple presentations), a deficit is observed. Thus, these findings revealed a common, albeit not universal, aspect of learning by the medial temporal system, that this system acquires information rapidly. Under the conditions where normal acquisition was rapid, animals without that system were disadvantaged. Conversely, under conditions where the rapid learning system conferred no advantage, no learning deficit was observed. This combination of observations, and several other findings, showed that many of the central features of the phenomenology of human amnesia can be modeled in animals, and specifically in monkeys. This work set the stage for a more detailed examination of which medial temporal lobe structures are critical for memory in monkeys; this work is discussed in Chapter 9. A parallel effort was also ongoing to model amnesia in other animals, particularly rats. The results of this effort are summarized next. Can declarative memory be modeled using rats? Scoville and Milner's 1957 report on H.M. had suggested that, among the structures damaged within the temporal lobe, damage to the hippocampus in particular was responsible for the memory deficit. Therefore, following the initial reports on human amnesia, several laboratories developed procedures for ablation of the hippocampus in rats, as well as cats and rabbits. As was the case with monkeys, the first tests to be employed in examining the effects of hippocampal damage were a variety of simple

114

Part II: Cognition

conditioning, discrimination learning, and maze learning tests that were the focus of current research by learning theorists. However, the results using each of these formal tests seemed inconsistent by any simple analysis. The findings were puzzling at best, and certainly did not support a conclusion that hippocampal damage results in severe and global amnesia in rats or other nonprimates. A few of these findings are summarized next to illustrate the confusing state of affairs that ensued from this research. The earliest assessments of learning and hippocampal function in rodents included two different tests of conditioning animals to avoid noxious stimuli (usually irritating electrical shocks). One of these, called "shuttle-box avoidance," involved training rats to alternate (shuttle) between two adjacent compartments of an alleyway. In one version of this task, each trial began with a buzzer that signaled the rat to shuttle to the alternate chamber before the floor in the currently occupied side was electrified by a mild current. The surprising result was that rats with large ablations of the hippocampus and overlying cortex learned the task in fewer trials than normal rats or rats with cortical damage only, and they retained this learning solidly across testing days. Shortly after, and contrasting with the first results, another study reported that rats with hippocampal ablations were unable to learn a different sort of avoidance task called "passive avoidance." Initially, hungry rats were trained to approach a chamber that had food. They began each trial in a large compartment, and were then signaled by a door opening to leave that compartment and to approach and enter the small food-containing chamber. After learning to execute the approach behavior immediately upon opening of the door, one day they were shocked while eating and driven out of the reward chamber. Normal rats and rats with hippocampal ablations rapidly learned the initial approach response and, consistent with the earlier experiment, rats with hippocampal ablations had shorter approach latencies and less variability in learning. However, the two groups responded quite differently in the avoidance component of training. After having been shocked just once in the reward chamber, none of the normal rats reentered again. By contrast, each of the rats with hippocampal damage did return to the chamber in which they were shocked. Although there was a learning impairment here following hippocampal damage, the overall pattern of findings did not offer compelling support for a simple amnesic disorder: After outperforming normal rats on the initial learning phase, rats with hippocampal damage showed a deficit in passive avoidance that seemed to reflect an inability to give up the previously acquired response, rather than a failure to learn per se. What a confusing state of affairs!

Exploring Declarative Memory Using Animal Models

115

Rats with hippocampal damage also performed perfectly normally, or better than normal rats, in the acquisition of standard Skinnerian conditioning tests in which rats learned to press a bar for food rewards. In addition, the pattern of early findings on the acquisition of simple sensory or spatial discriminations was no easier to reconcile with the deficit in human amnesia than were the findings on approach and avoidance learning. For example, in one of the first of these studies, rats with hippocampal ablations acquired at the same rate as normal animals a visual discrimination in which the stimuli were presented simultaneously. This task employed a Y-shaped maze composed of a start arm and two choice arms: one choice arm was black and the other white, and their left-right positions were randomly changed across trials (Fig. 5-3, left). The rat began each trial at one end of the start arm and then was rewarded for consistently selecting one color choice arm by food placed at the end of that arm. The opposite result was obtained in another version of the same visual discrimination task where the critical stimuli were presented successively on separate trials (Fig. 5-3, right). On each trial of this test, the rats were

Figure 5-3. A sequence of trials on two different versions of a visual discrimination task using a Y-maze.

116

Part II: Cognition

presented with one of two mazes, either a maze with two black goal arms or a maze with two white goal arms. The right arm contained food when the goal arms were white, whereas the left arm contained the food when the goal arms were black. In this variant of visual discrimination, rats with hippocampal damage required over twice as many trials as normal rats to reach a learning criterion. However, after a 2-week period, rats with hippocampal ablations showed as good retention as normal rats. As was the case with passive avoidance, although a discrimination learning impairment was observed following hippocampal damage, the overall pattern of results could not be interpreted as supporting a rodent model of global amnesia. Rats with hippocampal damage were not consistently impaired in learning a set of seemingly similar visual discriminations, and there was no impairment in long-term retention in either version of the task. Many subsequent efforts continued to provide mixed results on the effects of hippocampal damage on simple discrimination learning in rats and other species. There were some kinds of tasks where a disproportionate number of studies showed either impairment or spared learning: Discrimination learning was intact in three times as many experiments as not, although there were also many examples of deficient learning in each situation. This was the case across a broad variety of critical stimuli: nonspatial stimuli including specific visual, auditory, tactile, or olfactory cues, and spatial stimuli including left and right arms of a T- or Y-shaped maze. By contrast, a different result was obtained when animals were required to "reverse" the reward assignments, that is, when they were required to relearn the discrimination for same stimuli but each stimulus had the opposite reward assignment. In these situations animals with hippocampal damage more often showed impairments, although again there were many exceptions. Some researchers suggested that the deficit following hippocampal damage was an impairment in withholding previously learned responses. There was a constituency for this idea, but surely this was not the sort of conclusion that supported a straightforward model of amnesia in animals. Several new ideas suggested the hippocampus is involved in only one type of memory In the mid-1970s, several breakthroughs were made in the establishment of a rodent model of amnesia. Parallel to the successful approach in understanding human amnesia, a number of proposals emphasized the critical participation of hippocampus in some aspect of memory and lack of critical involvement in some other aspect of memory processing. Despite

Exploring Declarative Memory Using Animal Models

117

major differences in the fundamental processing function assigned to the hippocampus, all the proposals shared two general aspects of their formulations that are very important. First, each proposal espoused the view that the hippocampus plays a selective role in a distinct, higher-order form of memory, whereas hippocampal-independent mechanisms are sufficient to mediate simpler forms of learning and memory. The recognition of multiple forms of memory, only one of which depends on the hippocampus, constituted a major breakthrough of research on rodent memory in that period. Second, there was substantial agreement on the characteristics of the kind of learning that was successful mdependent of hippocampal function. All the proposals that emerged in this period described the capacities of animals with hippocampal damage in a manner consistent with characterizations of "habit" learning. Some characterized learning without the hippocampus as involving dispositions of specific stimuli into approach and avoidance categories, and as involving slow and incremental behavioral adaptation to the stimuli. Others characterized the behavior of animals with hippocampal damage as prone to rigidly adopt permanent assignments of cues and behaviors associated with reinforcement very early. The combination of these qualities remains undisputed in accounting for the success of rats with hippocampal damage in simple approach and avoidance conditioning and in discrimination learning. Beyond general agreement on what the hippocampus does not do, however, the theories differed substantially. In the following section two prominent views are discussed. Both of these proposals received substantial support from a particular key line of experimentation. At the same time each line of experiments challenged the conclusions from the other theory. A cognitive map in the hippocampus? In 1978 John O'Keefe and Lynn Nadel proposed that the hippocampus implements the cognitive maps described by Tolman. In a monumental achievement, O'Keefe and Nadel surveyed the extensive research findings on the anatomy and physiology of the hippocampus and on the studies of the effects of hippocampal damage in animals and humans. Each of these areas of knowledge was interpreted as supporting their overall hypothesis that the hippocampus is specifically dedicated to the construction and use of spatial maps of the environment. In their survey of the studies on animals with damage to the hippocampus or its connections, O'Keefe and Nadel emphasized that hippocampal damage typically results in severe impairment in most forms of spatial exploration and learning. Conversely,

118

Part II: Cognition

they noted that impairment of nonspatial learning (such as simple discrimination learning) is less commonly reported. These conclusions were combined with evidence of hippocampal "place cells," hippocampal neurons that signaled the location of the animal in space while it explored its environment (see Chapter 6 for details), leading them to suggest that hippocampal spatial information processing is a critical element of creating maps of space. It is important to emphasize that O'Keefe and Nadel's analysis went well beyond making a simple distinction between "spatial" and "nonspatial" learning. They proposed that the acquisition of cognitive maps involves a wholly distinct form of cognition from that of habit formation. Cognitive maps involve the representation of places in terms of distances and directions among items in the environment, and are composed as a rough topological map of the physical environment that the animal uses to navigate among salient locations and other important cues. They envisioned cognitive maps as enabling animals to act at a distance, that is, to navigate to locations beyond their immediate perception. In addition, cognitive mapping was characterized by a rapid, all-or-none assignment of cues to places within the spatial map. This kind of learning was envisioned as driven by curiosity, rather than reinforcement of specific behavioral responses, and as involving relatively little interference between items because they would be represented separately in a map or in different maps for distinct situations. In short, spatial mapping had most of the qualities of Tolman's cognitive maps, and therefore represented the form of cognitive memory as distinguished from habit learning. To get a feel for the striking and selective impairment in spatial learning following hippocampal damage in animals, consider the evidence from the water maze test, a spatial memory task that has received widespread use in studies of learning and memory. Originally developed by Richard Morris in 1981, this apparatus involves a large swimming pool filled with tepid water made murky by the addition of milk powder (Fig. 5-4A). An escape platform is hidden just beneath the surface of the water at an arbitrary location. Rats are very good swimmers and rapidly learn to locomote around the pool, but they prefer not to swim and will seek the platform so they can climb onto it. Animals cannot see the platform directly, but instead must use distant spatial cues that are visible above the walls of the pool around the room. On each training trial, the rat begins swimming from one of multiple locations at the periphery of the maze, so that it cannot consistently use a specific swimming course to reach the escape platform. Rats learn to use a spatial navigation strategy to find the platform even after training from a consistent starting

Exploring Declarative Memory Using Animal Models

119

Figure 5-4. The Morris water maze task. A: A sketch of the apparatus and two trials of a typical training sequence. In the place navigation version of the task, the rat begins each trial at one of four randomly selected locations and must find a submerged platform (dashed circle) positioned at a constant location. In the cued navigation version, the rat also starts at one of four locations, and the platform is visible (solid circle) and moved randomly across trials. B: Performance of rats with hippocampal lesions, cortical lesions, and normal controls in acquiring the water maze task. Place navigation = hidden platform; cue navigation = visible platform. C: Performance on the transfer test. Left: Swim path of a control subject; dashed lines indicate quadrant of the maze in which the platform had been located. Right: Swim times of rats in different maze quadrants; black bar corresponds to the training quadrant (data from Morris et al., 1982).

point, as evidenced in their ability after training to locate it efficiently from novel starting points. Morris and his colleagues showed that hippocampal ablation results in severe impairments in the water maze task (Fig. 5-4B). In the initial trials, all animals typically required 1-2 minutes to find the platform. Dur-

120

Part II: Cognition

ing the course of repeated trials, normal animals rapidly reduced their escape latency, such that they eventually reached it in less than 10 seconds from every starting point. Rats with hippocampal ablations also reduced their escape latencies, showing some extent of learning. However, they reached asymptotic performance at approximately 35 second latencies, largely due to a reduction of completely ineffective strategies such as trying to climb the walls; however, they never learned to swim directly to the platform location in the manner that normal rats do. In a subsequent "transfer test" the escape platform was removed and rats were allowed to swim for 1 minute with no opportunity for escape. In this transfer condition, normal rats circled in the close vicinity of the former location of the platform, as measured by a strong tendency to swim within the quadrant of the pool in which it had been located (Fig. 5-4C). Rats with hippocampal ablations showed no preference for the quadrant of the platform, highlighting the severity of their spatial memory deficit. In a different version of the water maze task, when the escape platform could be seen above the surface of the water (cue navigation), both normal rats and rats with hippocampal ablations rapidly learned to swim directly to it. This protocol emphasized the distinction between intact learning to approach the platform guided by a specific local cue, versus no capacity for learning guided by the relation among distant spatial cues. Since Morris and colleagues' original experiment, several studies have confirmed the selective impairment on the spatial version of the Morris water maze task following hippocampal damage, and this task has become a benchmark test of hippocampal function in rodents. An alternative theory: The hippocampus and remembering recent experiences In 1979, David Olton and his colleagues proposed an alternative view of the role the hippocampus plays in learning and memory. He argued that the hippocampus is critical when the solution of a problem requires memory for a particular recent experience. He called this "working memory," but note that we will not use this term because it has a different meaning in the current cognitive and neuroscience literatures. In current usage, working memory refers to the ability to temporarily hold information online while the subject is working on that information. As described in detail in Chapter 13, this memory capacity has been tied to the function of prefrontal cortex rather than hippocampus. The kind of memory Olton described involved the capacity to remember information that was obtained in a single experience, and to retain and then use it after any delay

Exploring Declarative Memory Using Animal Models

121

and over substantial interpolated material, exceeding the properties of working memory. Olton and his colleagues distinguished this kind of memory from "reference memory," which he characterized as memory for information that is constant across trials. Such reference information includes that there is a food reward at the end of each arm, that the rewards are not replaced during a trial but are replaced between trials, etc. On a conceptual level Olton viewed his theory as capturing the distinction raised by Tulving (1972) between "episodic memory" events tied to specific time and place, as contrasted with "semantic memory" for knowledge that is time- and event-independent (see Chapter 4). Accordingly, it is more appropriate to describe Olton's characterization of hippocampal memory as memory for unique episodes. Olton and colleagues' evidence was generated in great part from experiments using a novel test apparatus called the radial-arm maze. This maze is composed of several (typically eight) runway arms radiating outward like spokes of a wheel from a central platform (Fig. 5-5A), and there are many variants on the number of arms and reward contingencies involved in this task. In the standard version of the task, at the outset of each trial a food reward is placed at the end of each of the arms of the maze. During the course of a trial, the rat is free to enter each arm to retrieve the food rewards. Once retrieved, the food rewards are not replaced during that trial. Rats rapidly learn to approach each arm just once on a given trial. On subsequent trials, all of the arms are baited again just once. The number of arms entered during a trial provides a measure of the ability of the rat to remember which arms it has visited on that particular trial. Notably, this is not a test for memory of spatial locations—typically the spatial cues themselves are never hidden and do not need to be remembered. Instead, the central memory demand is to remember the animal's most recent visits to particular arms of the maze, that is, the recent behavioral episodes as opposed to the many other times each arm has been visited. In a classic experiment demonstrating the specificity of the impairment following hippocampal damage, Olton and his colleagues compared performance for different maze arms that had distinct reward contingencies (Fig. 5-5B). Some arms of the maze were baited once each trial, following the typical contingency that required memory for specific experiences. In addition, other arms were never baited, and rats were to learn across trials not to enter these arms at all. The latter capacity is another example of "reference memory" emphasizing the fixed nature of the stimulus-response associations for these arms. After initial training of all animals to high levels of performance on both components of the task, half

Figure 5-5. Three variations of the radial maze task. A: All maze arms are baited once, and the rat must visit each without a repetition. B: Only half the arms are baited once and the other half are never baited. Rats learn to visit each baited arm once per trial and never to visit the unbaked arms. C: The nonspatial version of the task. Each arm has a different surface on the floor, and the arms are interchanged randomly between trials. Rats learn to visit each cue once per trial.

122

Exploring Declarative Memory Using Animal Models

123

the animals had the hippocampus disconnected by transection of a major fiber bundle called the fornix. Subsequently in postoperative testing, normal animals continued to perform at very high levels on both the components of the task. By contrast, rats with hippocampal disconnections performed less well than normals from the outset of postoperative testing on both components of the task, but improved rapidly on their reference memory choices. However, they never performed better than chance even with extended training on the arms that required memory for specific experiences. Olton's proposal was seen as in direct conflict with the cognitive mapping notion, and many succeeding experiments provided support for one or the other proposal. Olton's hypothesis provided the better explanation of one of the early and persistent complexities of data on spatial discrimination, reversal learning, and alternation. Early studies found that hippocampal damage generally does not prevent learning of a simple spatial discrimination on a Y- or T-maze. O'Keefe and Nadel were quick to point out that this maze problem can be solved in two ways, by orienting for a left or right turn or by going to the place of reward. In their view, the hippocampus mediates the place strategy but not the orientation strategy, so that when deprived of hippocampal function the rats have an alternative solution and hence are unimpaired. However, an equally frequent result in the early studies is that rats with hippocampal damage are impaired at spatial reversal learning and are unable to learn to alternate left and right arm selections in Y- or T-mazes, and can show the clear dissociation between intact spatial discrimination and impaired alternation in the same T-maze apparatus outfitted with different types of choice points. It might well be that rats find it easy to adopt an orientation strategy in spatial discrimination, but strongly favor a place strategy in reversal learning and have to use a place strategy in delayed alternation. But it is not the nature of the spatial cues that differentiates these alternatives among these tasks. Rather, it is the demand to use the same cues in different ways for each task that is the critical factor. Olton's memory dichotomy accounts for these results more directly, in that only reference memory is required for the spatial choice never rewarded, but memory for the most recent experience is required for the alternation of choices. Another line of evidence strongly favoring Olton's theory consisted of experiments extending the role of the hippocampus to memory for specific experiences in the radial maze guided by nonspatial stimuli. The structure of the task was generally the same as in the previous tests of spatial working memory. Each of multiple maze arms was baited just once and perfect performance was measured as the ability of the rat to obtain all

124

Part II: Cognition

the rewards by entering each arm only one time. For the nonspatial version of this task, however, they created a four-arm maze in which the walls and floor of each arm were covered by a distinct set of tactile and visual cues, and distant cues were eliminated by covering the arms with a translucent gauze (Fig. 5-5C). Furthermore, after each arm entry the rat was briefly confined to the central platform while the arms were rearranged to eliminate any consistent arm positions or their configuration. Thus, the rats had to remember entered arms by their distinctive internal (intramaze) cues and ignore any spatial cues about the locations of the arms. After initial preoperative acquisition, normal rats continued to perform the task well but rats with fornix transections failed to reacquire the task even with extended retraining, despite the nonspatial nature of this variant of the task. The findings on these two theories of hippocampal function leave us in a quandary. O'Keefe and Nadel's theory predicts impairment on any task that requires the use of spatial cues in a cognitive map. It is not entirely clear that performance on the radial maze requires cognitive mapping. But the same spatial stimuli guide performance on both the episodic memory and reference memory versions of the task (Fig. 5-5B). So, the cognitive mapping view would not predict a difference in role of the hippocampus in the two versions of the task. On the other hand, the pattern of deficits on the radial maze tasks indicates that memory for specific prior episodes is critical, and not a demand for the use of spatial cues per se. At the same time, the Morris water maze task is clearly a "reference" memory task. Learning this task requires hippocampal function, a finding that cannot be explained by Olton's hypothesis. Clearly these two theories each capture a critical aspect of hippocampal functioning. However, another fundamental formulation must be pursued to account for both sets of findings. The following section diverges from these maze studies and reconsiders the properties of declarative memory that emerged from studies on human amnesia. Subsequently I return to the findings from the maze studies, as well as other experiments, and offer a reconciliation of the findings within the framework of properties of declarative memory. Convergence on the relational account of hippocampal function in memory As discussed in the preceding chapter, characterizations of memory functions of the medial temporal lobe in humans focus on declarative memory, the memory for everyday facts and events that can be brought to conscious recollection and can be expressed in a variety of venues. An

Exploring Declarative Memory Using Animal Models

125

approach to investigating this kind of memory can be obtained through a deeper consideration of the fundamental features of declarative memory. First, consider the notion that declarative memory is a combination of "event" or episodic memory and "fact" or semantic memory. How do these two kinds of memory combine to compose declarative memory? We acquire our declarative memories through everyday personal experiences, and in humans the ability to retain and recall these "episodic" memories is highly dependent on the hippocampus. But the full scope of hippocampal involvement also extends to semantic memory, the body of general knowledge about the world that is accrued from linking multiple experiences that share some of the same information. For example, a typical episodic memory might involve recalling the specific events and places surrounding the meeting of a long-lost cousin. Your general knowledge about the relationships of people that compose your family tree and other facts about the history of your family come in great part from a synthesis of the representations of many meetings with relatives and other episodes in which family personalities or events are observed or discussed. Similarly, our episodic memory mediates the capacity to remember a sequence of events, places passed, and turns taken while walking across a city, and a synthesis of many such representations provides general knowledge about the spatial layout of the city. Second, consider the nature of declarative memory expression. Declarative memory has been characterized as available to conscious recollection and subject to verbal reflection or other explicit means of expression. By contrast, procedural memory has been characterized as the nonconscious acquisition of a bias or adaptation that is typically revealed only by implicit or indirect measures of memory. Thus, declarative memory for both the episodic and semantic information is special in that one can access and express declarative memories via various routes and these memories can be used to solve novel problems by making inferences from memory. For example, even without ever explicitly studying your family tree and its history, one can infer indirect relationships, or the sequence of central events in the family history, from the set of episodic memories about your family. Similarly, without ever studying the map of a city, one can make navigational inferences from the synthesis of many episodic memories of previous routes taken. These descriptions present a formidable challenge for the study of declarative memory in animals. We do not have the means for identifying episodic memory or monitoring conscious recollection in animals; the very existence of consciousness in animals is a matter of debate. An assessment of verbal reflection is, of course, out of the question, and it is not other-

126

Part II: Cognition

wise obvious how to assess episodic and semantic memory or evaluate "explicit" memory expression in animals. However, these aspects of memory may in fact be accessible if we consider further characterizations that have been offered to distinguish declarative and procedural memory. To the extent that these descriptions do not rely on consciousness or verbal expression, they might be operationalized for experimental analysis in animals. To this end, in 1984 Cohen offered descriptions that could be helpful toward the goal of operationalizing fundamental properties of declarative memory. He suggested that "a declarative code permits the ability to compare and contrast information from different processes or processing systems; and it enables the ability to make inferences from and generalizations across facts derived from multiple processing sources. Such a common declarative code thereby provides the basis for access to facts acquired during the course of experiences and for conscious recollection of the learning experiences themselves" (p. 97, italics added). Conversely, procedural learning was characterized as the acquisition of specific skills, adaptations, and biases and that such "procedural knowledge is tied to and expressible only through activation of the particular processing structures or procedures engaged by the learning tasks" (p. 96). Two distinctions revealed in these characterizations have been employed during development of assessments of declarative and procedural memory that may be applicable to animal studies. First, declarative memory is distinguished by its role in comparing and contrasting distinct memories, whereas procedural memory involves the facilitation of particular routines for which no such comparisons are executed. Second, declarative memory is distinguished by its capacity to support inferential use of memories in novel situations, whereas procedural memory only supports alterations in performance that can be characterized as rerunning more smoothly the neural processes by which they were initially acquired. These distinctions, plus consideration of the nature of episodic and semantic memory as described in humans, can be extended to make contact with the broad literature on hippocampal function in animals, resulting in a proposal for the representational mechanisms that might underlie declarative memory. Based on a consideration of these characteristics of declarative memory, Eichenbaum and Cohen suggested that the hippocampal system supports a relational representation of memories. Furthermore, a critical property of the hippocampal-dependent memory system is its representational flexibility, a quality that permits inferential use of memories in novel situations. According to this view, the hippocampal system mediates the organization of memories into what may be thought of as a "memory space."

Exploring Declarative Memory Using Animal Models

127

This theory has been elaborated to make contact with several of the key observations of the theories of memory described earlier. Within the relational memory theory, the memory space is constructed by an interleaving of episodic memories into a semantic structure in which memories are connected by their common elements. In this scheme the major components of a memory space are the representations of memories as sequences of events that compose specific experiences, that is, distinct episodic memories. Episodic memories, then, are interleaved into the memory space by shared events within related memories. Thus, the construction of family trees and layouts of cities are built up from linking together many episodic memories for specific encounters with family members and for specific trips through a city. Furthermore, such a memory space supports the kind of comparing and contrasting among memories that allows flexible and inferential use of memories. Within the memory space scheme these capacities are generated by the structure of the relational representation. When one element of the network is activated by a retrieval cue, all memories that contain that item and are sufficiently strongly associated will be activated, including elements of those memories that are elements of yet other related episodic representations. Consequently, memories that are only indirectly associated with the originally activated element would also be activated. Such a process would support the recovery of memories in a variety of contexts outside the learning situation and would permit the expression of memories via various pathways of behavioral output. Conversely, according to the relational memory account, hippocampalzwdependent memories involve individual representations, such memories are isolated in that they are encoded only within the brain modules in which perceptual or motor processing is engaged during learning. These individual representations are inflexible in that they can be revealed only through reactivation of those modules within the restrictive range of stimuli and situations in which the original learning occurred. One might expect individual representations to support the acquisition of task procedures that are performed habitually across training trials. Individual representations should also support the acquisition of specific information that does not require comparison and consequent relational representation. The combination of relational representation (a consequence of processing comparisons among memories) and representational flexibility (a quality of relational representation that permits inferential expression of memories) suggests an information processing scheme that might underlie declarative memory in humans and animals as well. Most important, this description of the nature of declarative memory is testable in animals.

128

Part II: Cognition

Testing the relational memory theory Large-scale networks for family trees and city layouts are but two examples of the kind of memory space proposed to be mediated by the hippocampus. Within this view, a broad range of such networks can be created, with their central organizing principle the linkage of episodic memories by their common events and places, and a consequent capacity to move among related memories within the network. These properties of declarative memory suggest an approach for the development of animal models. Thus, a way to study the development of a memory space from overlapping experiences, and to make inferences from the network knowledge, is to train subjects on multiple distinct experiences that share common elements and then test whether these experiences have been linked in memory to solve new problems. One can conceive of this approach as applied to various domains relevant to the lives of animals, from knowledge about spatial relations among stimuli in an environment, to categorizations of foods, to learned organizations of odor or visual stimuli or social relationships. In the remainder of this section some of the evidence supporting the relational account of hippocampal memory function is elaborated. The first set of experiments reexamines discrimination learning, showing once more that learning performance may be severely impaired or completely intact depending on performance demands, showing how a critical demand for relational processing leads to different behavioral outcomes. The second set of experiments examines directly the role of the hippocampus in flexible memory representations and the expression of memory by novel uses of previously acquired knowledge. The importance of linking multiple distinct experiences

A critical aspect of the relational theory is the interleaving of multiple experiences that share information into a larger network of memories. Therefore, it can be expected that the hippocampus will play a critical role in learning in situations where the task has a strong demand to synthesize multiple overlapping experiences. One such case involves spatial learning, similar to the example of the learning of routes through a city, but involving rats and the Morris water maze task. As described before, in the conventional version of this task, rats learn to escape from submersion in a pool by swimming toward a platform located just underneath the surface. Importantly, training in the conventional version of the task involves an intermixing of four different kinds of trial episodes that differ in the starting point of the swim (see Fig. 5-4A). Focusing on this task demand,

Exploring Declarative Memory Using Animal Models

129

the relational memory account offers a straightforward accounting of the pattern of deficits and situations where intact spatial learning is observed. Releasing the rats into the water at different starting points on successive trials strongly encourages the subjects to compare their views along the swim paths as they pass the positions of extramaze stimuli, forcing the animal to consider its relation to the positions of the cues across trials. Indeed, it is difficult to imagine how the task could be solved without synthesizing the information acquired during the different swim episodes into a representation of spatial relations among cues, allowing them to disentangle otherwise conflicting associations of the separate views seen from each starting point. Under these conditions of strong demands to interleave the different types of episodes, animals with hippocampal damage typically fail to acquire the task. The importance of the demand for interleaving four different types of trial episodes was demonstrated in an experiment that explored acquisition of the water maze when this requirement was eliminated. This experiment used a version of the task where rats were released into the maze from a constant start position on each trial. Initially, animals were trained to approach a visible black-and-white striped platform. Then the visibility of the platform was gradually diminished using a series of training stages that involved a large, visible, white platform, then smaller platforms, and finally sinking the platform below the water surface. With this gradual training procedure, rats with fornix transections learned to approach the platform directly, although they were slower to acquire the response at each phase of training (Fig. 5-6A). In addition, their final escape latencies were slightly higher than those of intact rats, due to an increased tendency for "near misses," trials on which they passed nearby the platform without touching it and forcing them to circle back. But, in contrast to the standard version of this task, animals with hippocampal damage were able to learn the location of the escape platform. Both sets of rats were using the same extramaze cues to guide performance, as indicated by the results from the standard "transfer" test in which the escape platform is removed and the swimming pattern of the rats is observed for a fixed period. Both normal rats and rats with hippocampal damage swam near the former location of the platform, indicating that they could identify the place of escape by the same set of available extramaze cues rather than solely by the approach trajectory. After several probe tests, these rats were to learn a novel escape location using multiple starting points. Rats with hippocampal damage failed completely. The success of rats with hippocampal damage on the constant start version of the task, contrasted with their failure on the standard, variable

130

Part II: Cognition

Figure 5-6. Place learning in the water maze. A: In the constant start version of this task, the rat always begins from the same location near the escape platform. Note that rats with hippocampal damage (fornix lesions) learn the task more slowly than normal rats, but eventually succeed. B: Navigation from novel starting locations. Six novel starting locations were used in the probe testing trials that were intermixed among repetitions of the instruction trial. C: Example swim paths for individual normal control rats and rats with hippocampal damage on the probe trial that began from the "east" start location (see black rat in B) (data fom Eichenbaum et al., 1990).

start version, indicating that it was not the use of distal spatial cues per se, but rather other factors governing how these cues were used that determined the critical involvement of the hippocampus. Rather, the cognitive demand that invoked critical hippocampal function was the requirement to interleave information from multiple experiences on different escape paths taken across trials.

Exploring Declarative Memory Using Animal Models

131

The flexible and inferential expression of spatial memories

Were there differences in the way in which rats with hippocampal damage and normal rats learned and represented the task? This question was addressed by using a series of probe tests, each involving an alteration of the cues or starting points, intermixed within a series of repetitions of the instruction trial. One of the probe tests demonstrated a particularly striking dissociation between the two groups of rats. In this test, the platform was left in its normal place but the start position was moved to various novel locations. When the start position was the same as that used during instruction trials, both normal rats and rats with hippocampal damage had short escape latencies (Fig. 5-6B). On the critical probe trials with novel starting positions, normal rats also swam directly to the platform regardless of the starting position. By contrast, rats with hippocampal damage rarely swam directly to the escape platform and sometimes went far astray, subsequently having abnormally long average escape latencies on these probe trials. This striking deficit in rats with hippocampal damage was demonstrated by a close examination of their individual swim trajectories (Fig. 5-6C). All the normal rats nearly always swam directly to the platform regardless of their starting point. But rats with hippocampal damage swam in various directions, occasionally leading them straight to the platform, but more often in the wrong direction, and they sometimes never found the platform in this highly familiar environment. The observation of a severe deficit in using spatial information acquired successfully to navigate from novel starting points constitutes strong evidence indicating the importance of the hippocampus in the flexible and inferential expression of spatial memories. Extending the role of the hippocampus in relational representation and representational flexibility to nonspatial learning and memory The preceding set of experiments provides compelling evidence that the hippocampus plays an important role in spatial learning by supporting the interleaving of multiple overlapping experiences and in using the resulting organized spatial representation to navigate from new locations. Now we consider whether this accounting applies globally to nonspatial as well as spatial memory organizations. One study that examined this issue directly explored the role of the hippocampus in learning an organization of odor stimuli, and in flexible and inferential expression of this organization. This study compared normal rats and rats with selective damage to the hippocampus on their ability to learn a set of odor problems and to inter-

132

Part II: Cognition

leave their representations of these problems to support novel inferential judgments about them. To accomplish this an odor-guided version of the so-called paired associate task was developed for rodents, and this task extended the learning requirement to include multiple stimulus-stimulus associations with overlapping stimulus elements (Fig. 5-7). The task was especially designed to take advantage of the superb abilities of rats to learn about odors, and to exploit their natural food foraging behaviors. The stimuli were common household spices, such as oregano, garlic, etc. These stimuli were mixed into ordinary playground sand and presented in small plastic cups. Rewards were a highly preferred sweetened cereal (Froot Loops) buried under the sand. Prior to formal testing the animals were exposed to cups of sand with buried cereal, and rapidly learned to forage through the sand to find the rewards. Animals were initially trained to associate pairs of odor stimuli with one another. For brevity, the initial pairs will be called A-B and X-Y, where each letter corresponds to a different odor. Each trial was composed of an initial presentation of one of two sample stimuli, A or X. Then that stimulus was removed and the pair of choice odors, B and Y, was presented (Fig. 5-7A). The rule for the choice was as follows: If A was the sample, then B should be selected to obtain a reward; if X was presented, Y contained the reward. Animals were trained on repetitions of these two types of trials (A-B and X-Y) until they achieved a criterion of 80% correct choices. Then they were trained on a second set of paired associates, and this time each association involved an element that overlapped with one of those in the previous pairings, B-C and Y-Z. So now the sample stimuli were B or Y, and the choice stimuli were always C and Z. This problem was also trained to the 80% correct criterion. Then they were trained with all four problems (A-B, X-Y, B-C, and Y-Z) intermixed (Fig. 5-7B). Normal rats learned each set of paired associates rapidly, and hippocampal damage did not affect acquisition rate on either of the two training sets or their combination. It is important to recognize that correct response for each pairing can be learned independently for each problem. That is, unlike the situation with the water maze described before, the solution to each problem does not interfere with solutions to any of the other problems. Therefore, it was not expected that damage to the hippocampus would affect the ability to acquire these problems. However, it was expected that normal rats and rats with hippocampus damage would form qualitative different representations of the problems, and this was examined next. Following successful acquisition of all the paired associates, subjects were given probe tests to determine whether they organized their repre-

Exploring Declarative Memory Using Animal Models

133

Figure 5-7. Paired associate task. A: Illustration of a rat performing the sample and choice trials of the task. B: Odors presented in the sample and choice trials in each training set (left) and performance of rats in learning each set. C: Performance on the probe tests for transitivity and symmetry. The preference index was calculated as the ratio of the difference in times spent (transitivity) or choice (symmetry) in digging between the two cups and that for the sum on the same measure for the two cups (data from Bunsey and Eichenbaum, 1996).

sentations of the four odor pairs into an efficient scheme that interleaved overlapping odors pairs, or simply learned each problem independently. In the preceding scheme, representations of the pairs A-B and B-C could be organized into a larger network, A-B-C, and similarly X-Y and Y-Z could be organized into X-Y-Z. Two kinds of probe tests were presented

134

Part II: Cognition

to examine whether these larger network representations were formed, and to assess their flexibility. One test assessed the capacity for "transitive inference," the ability to make a judgment about two odors that are only indirectly related by another, not presented item. In this scheme, note that odor A and odor C were never presented together but can be indirectly related via their shared associate B. Similarly, X and Z are indirectly related via odor Y. In a critical test for transitivity, subjects were presented with trials using the same sample-choice format with trials on A-C and X-Z. However, in these tests no reward was provided, and the time spent foraging in the two choice cups was compared as a measure of their recognition of the appropriate relations between indirectly associated elements. Normal rats also showed strong transitivity, reflected in a preference for odors indirectly associated with the presented sample (Fig. 5-7C). By contrast, rats with selective hippocampal lesions were severely impaired, showing no evidence of transitivity. The other probe involved a test for "symmetry" of the learned associations, the ability to recognize related pairings regardless of the order in which the items are presented. In this scheme, the subjects were asked to recognize appropriate pairings in the reverse order of that used in training on the second set of problems. Again, the sample-choice format was used, but the pairings were C-B and Z-Y. In the symmetry test, normal rats again showed the appropriate preference in the direction of the symmetrical association. By contrast, rats with hippocampal lesions again were severely impaired, showing no significant capacity for symmetry (Fig. 5-7C). The combined findings show that rats with hippocampal damage can learn even complex associations, such as those in the odor paired associates. But unlike normal animals, they do not interleave the distinct experiences according to their overlapping elements to form a larger network representation that supports inferential and flexible expression of their memories. A case of naturalistic learning, and natural inferential memory expression Extending the range of our study of hippocampal involvement in associative learning, the role of the hippocampal region has also been assessed using a type of social olfactory learning and memory, called the social transmission of food preferences. This test is based on observations of rat social behavior in the wild. Rats form large social communities and send out "foragers" to explore for food sources. When forager rats return to the nest, they inform others of new foods by carrying the odor of a new

Exploring Declarative Memory Using Animal Models

135

food on their breath. Subsequently, the other rats to whom this information is conveyed seek out that same food. This form of social learning is interpreted within the heuristic that a food recently consumed by the forager is safe, and thus transmitting this information is adaptive in rat social groups. This behavior can be easily adapted to a formal laboratory test by designating "demonstrator" (forager) and "observer" (subject) rats, and reproducing the elements of the foraging and social communication events (Fig. 5-8). Initially, the demonstrator rat is given rat chow tainted with a food spice, such as cinnamon or chocolate powder. Then the demonstrator is placed in the home cage of the subject for several minutes, during

Figure 5-8. The social transmission of food preferences task. Top: Protocol for the task. Bottom: Performance of intact rats (open circles) and rats with selective hippocampal lesions (filled circles) in immediate and 1-day retention tests (data from Bunsey and Eichenbaum, 1995).

136

Part II: Cognition

which the two animals socially interact. Then the demonstrator is removed, and the subject must remember the food odor for a variable period. Finally, the subject is presented with two food cups, one with rat chow tainted with the same odor as presented to the demonstrator and the other cup with another food odor that may be equally familiar or novel to the subject. Learning is demonstrated as an increase in the probability that the observer will later select that same food as the demonstrator had eaten over other foods. Previous investigations of this behavior in normal rats has shown that the mechanism underlying this learning involves an association between two odors present in the demonstrator rat's breath, the odor of the recently eaten food and a natural odorous constituent of rat's breath, carbon disulfide. These studies have shown that exposing the subject to the distinctive food odor alone, or to carbon disulfide alone, have no effect on later food preference. However, exposure to the combination of these two odors using a "surrogate rat," a cotton ball saturated with carbon disulfide, substitutes well for the social interaction in producing the appropriate learned food preference. Thus, the shift in food choice cannot be attributed to mere familiarity with the food odor. Rather, the conclusion from these studies is that the formation of a specific association between the food odor and carbon disulfide, in the absence of any primary reinforcement, is both necessary and sufficient to support the shift in food selection. Social transmission of food preferences provides a strong example of declarative-like learning in the natural behavior of rats. Learning involves the formation of a specific stimulus-stimulus association in a single training episode. Furthermore, this paradigm provides one of the best examples of a natural form of "inferential" expression of the memory. Thus, note that the training experience involves a social encounter without any feeding, during which the subject is exposed to an arbitrary stimulus and another natural stimulus that acts as a reinforcer. By contrast, the memory testing situation is nonsocial, and involves food selection choices without the presence of the same stimulus that reinforced learning the new food odor. This aspect of the task, expression of memory in a situation very different from the learning event, is strongly consistent with the declarative property of representational flexibility. The role of the hippocampus has been investigated using this task, assessing both immediate memory and delayed memory for social exposure to the odor of a novel food. The behavior of rats with hippocampal damage during the social encounter is entirely normal. In the memory test, normal rats showed a strong selection preference for the trained food odor,

Exploring Declarative Memory Using Animal Models

137

and this memory was robust when tested immediately and after a 1-day retention period (see Fig. 5-8). By contrast, rats with selective damage to the hippocampus showed intact immediate memory, but their performance fell to chance level of food choices within 24 hours. The observation of intact immediate memory, similar to the pattern of spared short-term memory in human amnesics, indicates that the hippocampus is not required for perceptual, motivational, or behavioral components of the social learning or the ability to express the learned food choice preferences. But the hippocampus is required for long-term expression of the memory in a situation that is far outside the learning context. Summing up There is a long and mixed history of attempts to model amnesia in animals. Efforts in both monkeys and rats were largely unsuccessful owing to the poor choice of memory tests and the lack of realization that there are different forms of memory, only one of which depends on the medial temporal lobe. However, there have now been many successful demonstrations of the characteristics of human amnesia in both monkeys and rats. In monkeys, major advances were made with the advent of the delayed nonmatch to sample test that provided a novel test of recognition memory. The results using this test plus those of a set of related tests showed that medial temporal lobe damage similar to that in H.M. results in a remarkably similar pattern of impaired and spared memory abilities. In rats, major advances were made following the realization that much of the ambiguity in the pattern of effects of hippocampal damage could be explained by distinctions between impairments in spatial memory versus spared nonspatial memory, and between memory-specific recent experiences versus spared learning guided by habits or dispositions of specific stimuli. However, experiments in support of each of these two accounts are in apparent conflict. The combined results from both these accounts, and across many other experiments, provide compelling evidence for a comprehensive account of the cognitive mechanisms of declarative memory. Various kinds of learning, spatial and nonspatial, simple and complex, can be accomplished independent of the hippocampus in animals, as indeed is the case in human amnesic patients as well. However, the hippocampus is required to link together the representations of overlapping experiences into a relational representation, and supports the flexible and inferential expression of indirect associations among items within the larger organization of linked memories. This hippocampal function applies across many situations, in-

138

Part II: Cognition

eluding navigation within spatial organizations, and nonspatial organizations of specific stimuli (e.g., odors) in logical schemes or in natural social behavior. The next chapter explores the nature of the neural code that accomplishes this networking function of the hippocampus.

READINGS Cohen, N.J. 1984. Preserved learning capacity in amnesia: Evidence for multiple memory systems. In The Neuropsychology of Memory, N. Butters, and L.R. Squire (Eds.). New York: Guilford Press, pp. 83-103. Cohen, N.J., and Eichenbaum, H. 1993. Memory, Amnesia, and the Hippocampal System. Cambridge: MIT Press. Eichenbaum, H. 1997. Declarative memory: Insights from cognitive neurobiology. Annu. Rev. Psychol. 48:547-572. Mishkin, M,, and Petri, H.L. 1984. Memories and habits: Some implications for the analysis of learning and retention. In The Neuropsychology of Memory, N. Butters, and L.R. Squire (Eds.). New York: Guilford Press, pp. 287-296. Morris, R.G.M., Garrud, P., Rawlins, J.P, and O'Keefe, J. 1982. Place navigation impaired in rats with hippocampal lesions. Nature 297:681-683. O'Keefe, J.A., and Nadel, L. 1978. The Hippocampus as a Cognitive Map. New York: Oxford University Press. Olton, D.S., Becker, J.T., and Handlemann, G.E. 1979. Hippocampus, space, and memory. Brain Behav. Sci. 2:313-365. Squire, L. 1992. Memory and the hippocampus: A synthesis from findings with rats, monkeys, and humans. Psycho. Rev. 99(2): 195-231.

6 Windows into the Workings of Memory

STUDY QUESTIONS What is brain imaging, and how can it be used to study human memory? What are the characteristics of memory processing that activate the medial temporal lobe in normal human subjects? What are place cells? What is the full characterization of events that are encoded by hippocampal neuron firing patterns? What is a "memory space," and how might it be constructed by the properties of hippocampal neuron firing patterns?

S

o far this review of our understanding of hippocampal function in memory has focused predominantly on studies of amnesia in humans and on the effects of hippocampal damage in animals. In the present chapter, complementary evidence is presented from other related approaches. These approaches involve monitoring the ongoing operation of the human hippocampus and related brain structures during memory performance, providing a virtual "window" into the inner workings of the normal brain. This is accomplished at two levels of analysis: by using functional neuroimaging methods in normal humans and by recording the activity patterns of single neurons in animals. With regard to the functional imaging studies, there are now multiple sophisticated methods that are employed, particularly positron emission tomography (PET) and functional magnetic resonance imaging (fMRI). It 39

140

Part II: Cognition

is beyond the scope of this text to discuss the details of these methodologies, except to say that both involve measurements of blood flow and brain oxygen consumption, which provide good reflections of the level of activation of a brain area encompassing several thousand neurons over a second-to-second time scale. By telling us when, for example, the medial temporal area becomes active, functional neuroimaging studies in humans can inform us about the aspects or kinds of memory in which the hippocampal system is and is not involved, providing a way to assess its functional role. Single cell recording studies in animals provide an even closer look at the inner workings of the hippocampus. This method involves monitoring the action potentials of individual neurons and so allows a major increase in resolution of cellular activity within different parts of the medial temporal area, and even allows us to distinguish particular types of neurons within a specific brain structure. Also, this method has a greater resolution in time, allowing us to capture millisecond-to-millisecond computations by the fundamental elements of neural processing. Thus, these two approaches have complementary strengths and limitations. The strength of functional imaging is that it allows the simultaneous examination of the entire system, but at only a gross level that tells us which structures are activated to major shifts in task demands. Single cell recording methods allow us to monitor only one part of the system at any time, but offer insights into the fundamental coding properties of the units of neural computation. Functional neuroimaging studies of the human hippocampal system Early attempts to observe hippocampal activation in PET or fMRI during one or another memory performance were largely disappointing. Scannning during the study of materials such as word lists, and other tasks for which memory depends on hippocampal function, failed to show increased activation over nonmemory processing of the same materials. However, this turns out to be an artifact of the procedures in analysis of brain images. All imaging techniques involve a comparison of activation levels between two conditions, an experimental and a control condition. In memory studies, the experimental condition involves the critical memory demand under study, for example, memorizing word lists. The control condition involves the same perceptual and cognitive demands, except without the critical memory demand. In the word list example, the control might be reading words without having to remember them. The areas associated

Windows into the Workings of Memory

141

with memory processing per se are defined operationally as those brain regions where the activation level is greater in the experimental condition than in the control condition. However, the evidence described previously, as well as other physiological data that are described later, indicate that the hippocampus is always active in encoding new information for declarative memory. In the example, the control condition where subjects merely read words may invoke quite substantial hippocampal activation associated with remembering the experience of reading the word list. More generally, there is virtually no condition where memory processing is expected to be altogether absent. Thus, the failure to find activation associated with memory in the early studies may be attributed to the lack of a substantial difference in the activations associated during the memory condition (e.g., studying a word list) and apparent nonmemory activities for which memory processing was nevertheless invoked (reading a word list). Since those early studies, another generation of experiments has taken the issue of control conditions into consideration and in recent years has shown that the human hippocampal system can be seen in action during various memory performances and the results of these studies correspond well to those of the studies of amnesia. Thus, the findings from functional brain imaging generally support the distinction between declarative and procedural memory, and they provide new details on the nature of events that activate the medial temporal area. The following section explores these characteristics of medial temporal activation. The medial temporal area is involved in "global" information processing Consider first the range of to-be-remembered materials over which the medial temporal area operates. Studies of patients with bilateral damage to the medial temporal area have shown that amnesia is a global memory deficit. As discussed in Chapter 4, amnesic patients have a memory impairment that crosses between different learning materials and sensory modalities, encompassing verbal and nonverbal, spatial and nonspatial materials, regardless of whether they are presented visually, auditorily, etc., indicating that the role of this region in memory is, likewise, nonspecific with regard to material and modality. However, other studies involving patients with unilateral damage, that is, involving damage only to the left or right medial temporal lobe region, have shown clear material-specific memory impairments: Verbal memory performance is selectively compromised after left medial temporal lobe damage, and nonverbal memory performance is selectively compromised after right temporal lobe damage. In

142

Part II: Cognition

other words, there is a laterality in the critical medial temporal lobe contribution to memory, corresponding to the different types of learning materials. Furthermore, the nature of processing for which the hemispheres are specialized follows the well-known laterality for nonmemory processing of verbal versus nonverbal materials assigned to the left and right hemispheres, respectively. How well do the findings from functional imaging studies of memory correspond to this picture from studies of amnesia? Across a variety of functional imaging studies one can see both the globalness of medial temporal processing, considering both hemispheres, and also the material-specificity of left versus right hemisphere processing. Looking across the range of studies that have reported medial temporal activation, one sees that a range of stimulus materials can engage this system. Thus, ignoring the particular hemisphere in which activation occurs, activation in the medial temporal lobe has been reported for words, objects, scenes, faces, and spatial routes, landmarks, or locations. In addition, in studies that compared different classes of materials, clear hemispheric specialization has been seen. The results generally indicate greater left than right activation for words, and greater right than left activation for novel faces or objects. Accordingly, with regard to the scope of the materials processed by the medial temporal lobe, there is good concordance of the functional imaging and the data from studies on amnesia. Next we consider several general findings on the types of memory processing that have consistently activated the medial temporal area across multiple studies. These include activation by the simple presentation of novel stimulus information, activation by the processing of new stimulus associations, and activation associated with explicit, conscious recollection. Activation of the medial temporal region by the presentation of novel information There are now several lines of evidence that have suggested medial temporal lobe involvement in specific aspects of cognitive processing, although some of these are not identical to dimensions that are featured in accounts of amnesia. However, it is important to keep in mind that functional imaging studies are likely to reveal the initial stages of memory processing when the medial temporal area becomes involved, even in situations where its involvement may not be critical to that particular type of processing. A good example involves several reports of engagement of the medial temporal region by the mere presentation of novel information. These studies have found greater activation for the processing of novel as compared to familiar pictures of complex scenes or objects. In one study, subjects viewed

Windows into the Workings of Memory

143

magazine photos of scenes in alternating blocks of trials that included either a series of different scenes, each presented once, or just one scene that was presented repeatedly as a control. Subjects were instructed to study the scenes so they might be able to recognize them later, and they were scanned during this study phase. Greater activation was seen in the medial temporal region for the novel scenes in the experimental condition compared to the single repeated scene in the control condition. In perhaps the most striking example of a novelty detection effect, the right hippocampus was selectively activated by the presentation of novel visual noise patterns, that is, a series of random dot patterns, compared to the presentation of the same patterns a second time (Fig. 6-1 A). These findings have led some investigators to propose that the medial temporal region may be involved in novelty detection per se. However, as you will see later, the medial temporal region is activated also by familiar material under circumstances where the subject is making recognition judgments. So, clearly the detection of novelty per se is not the primary function revealed in activation by novel stimuli. Rather, an effect of stimulus novelty on medial temporal activation makes good sense as a reflection of the kind of processing that would underlie declarative memories of the many pictures. The presentation of a brand new scene every few seconds should invoke considerable declarative memory processing as the hippocampus is involved in encoding the information within scenes and the sequence of constantly changing scenes. By contrast, the control condition that involves repeated presentations of a single unchanging scene, minimizes this processing demand. Therefore, the comparison of these two conditions is likely to reveal the differential activation for maximal and minimal processing of new information. Medial temporal activation associated with processing stimulus associations The data just described suggest that the hippocampus always becomes engaged when new material is presented, but may become more activated to the extent that processing relations among elements within or across scenes is strongly demanded by the materials or the task. This view could account for why the early studies failed to find medial temporal activation when subjects were presented with highly familiar stimuli such as word lists, as compared to other manipulations of the same materials that also invoke memory processing. The studies that focused on the presentation of several changing scenes as compared to the same unchanging scene may have revealed a difference in the amount of hippocampal activation de-

144

Part II: Cognition

Figure 6-1. Functional imaging of the human hippocampus. A: Activation of the right hippocampus during viewing of novel as contrasted with familiar visual noise patterns (from Martin, 1999). B: Activation of the left hippocampus during recall of memories that are both temporally specific and personally relevant (from Maguire and Mummery, 1999).

manded by the differential requirement for processing information within and across items that would be encoded in memory. Consistent with this expectation the strongest support for the medial temporal region activation in processing information across scenes for explicit memory conies from a study by Henke and colleagues in which subjects were shown a series of pictures, each of which showed a person and a house (either the interior or the exterior) simultaneously. Subjects were

Windows into the Workings of Memory

145

required to judge if the person was likely to be the inhabitant or a visitor to that house, based on their appearance and that of the house. For example, one pair of pictures showed an elegant lady and an equally elegant sitting room that would constitute an appropriate match. Another example showed a disheveled man and a large mansion, representing an unlikely match. Thus, while the task did not directly demand the subjects to associate the two images, the nature of the judgment encouraged them to make an association between the person and the house. As a control, subjects were shown other pairs of pictures and requested to make separate decisions about the person's gender (male or female?) and view of the house (exterior or interior?). This test requirement encouraged the subjects to encode the house and person separately. Greater medial temporal activation was observed when the subjects were encouraged to associate the person and the house than when they encoded the same kinds of items separately. These findings suggest that it is not merely the processing of novel pictures that activates the hippocampus. If this were so, one would expect the same level of activation for similarly novel pictures of people and houses. Instead, the findings support the view that the medial temporal region is more activated when there is a greater demand for learning associations or relations between items in single learning episodes, as compared with processing the items separately. In a related study, Maguire and Mummery examined the activation of medial temporal structures during recollection of different types of materials. They found a large network of brain areas activated during different types of recall associated with different aspects of real-world memory, including whether the memory occurred at a particular time or instead involved time-independent factual information, or was personally relevant or instead involved general knowledge of public events. They found a striking and specific activation of the left hippocampus when what subjects recollected was a combination of personally relevant and specific in time (Fig. 6-1B). Medial temporal activation associated with conscious recollection More directly related to declarative memory, other studies have considered whether the medial temporal lobe region is disproportionately engaged in explicit memory, that is, in conscious recollection. In one experiment, subjects first studied word lists and then their memory was later tested in different ways during scanning. At testing, subjects were given

146

Part II: Cognition

word stems and asked to complete the stems either with the first word that came to mind, just as in a typical repetition priming experiment (see Chapter 4), or with a word from the study list, to engage conscious recollection, or with the first word that came to mind that was not on the study list, as a baseline condition where the subject stared at a cross-hair on the screen. Medial temporal activation was found for the condition that involved conscious recollection, as compared to either the priming condition, which involved implicit memory instructions, or the baseline condition. Other studies have related medial temporal activation to successful retrieval of previously stored information, that is, success in the effortful reactivation of stored representations. In one such study, subjects listened to two lists of words prior to being in the scanner. For one list, subjects were to decide whether each word was said by a male or female speaker. This condition was employed to encourage processing at the level of perceptual features of the words and to deemphasize encoding the semantic or conceptual content in the words. For the other list, the subjects had to decide whether each word referred to a living or nonliving thing, in order to encourage conceptual or semantic encoding over perceptual encoding. Subjects were subsequently tested for recognition memory while being scanned, in a series of test blocks that assessed memory separately for perceptually encoded and semantically encoded words. Greater medial temporal lobe activation was observed for test blocks that involved words from the semantically encoded list compared to the perceptually encoded list. The fact that there was a higher rate of successful recall of semantically encoded words compared to perceptually encoded words suggested that increased medial temporal activation for semantically encoded words was a consequence of the role of this region in successful recall, that is, in successfully gaining access to some memory representation. This connection was seen more formally as a strong positive correlation across subjects and conditions between test performance and medial temporal lobe activation. Another line of studies has shown that the medial temporal region is activated when subjects are involved in effortful retrieval and manipulation of retrieved geographical information. In these studies experienced London taxi drivers were scanned as they were requested to retrieve information about routes around London. Greater activation of medial temporal structures was seen during recall of route information than during recall of famous landmarks, movie plot lines, or movie scenes. In another study subjects learned to navigate around in a virtual reality environment. Subjects were then scanned while they made judgments about the ap-

Windows into the Workings of Memory

147

pearance or relative position of particular places in the environment compared to a control condition involving scrambled versions of the same stimuli. Medial temporal activation was observed bilaterally for both conditions that tested memory of the learned places compared to the control condition. These functional brain imaging studies indicate that the medial temporal region becomes activated whenever new, complex material is presented. Furthermore, these results indicate that the medial temporal region is maximally activated when processing this material demands encoding new associations or relations among separate items in specific memory episodes. In addition, the medial temporal area is similarly invoked during the act of explicit memory retrieval, that is, during conscious recollection. These findings are entirely consistent with the observations on the pattern of impaired and spared capacities in human amnesia, and thus demonstrate a close correspondence of the data from two different approaches. Furthermore, the combination of activations during both episodic learning and conscious recollection indicates that the medial temporal area is involved similarly in encoding and retrieval phases of declarative memory. The representation of experience in the activity of networks of hippocampal neurons The foregoing characterizations of the effects of damage to the hippocampal region (in Chapters 4 and 5), and of activation of the hippocampal region in humans, suggest that its fundamental role is in encoding rich episodic information and conscious and flexible memory expression. What kind of neural representation within the hippocampus would support this functional role? As described in the preceding chapter, on a conceptual level, the form of such a representation might be constituted as a large network that encodes episodic memories and links these memories via their shared features. Can such a scheme be confirmed and elaborated by observations on the elements of the hippocampus, that is, in the firing patterns of single hippocampal neurons? While any conclusions about the nature of firing patterns in the hippocampus is still quite preliminary, there is an emerging body of data consistent with this scheme. A wealth of evidence indicates that hippocampal neurons encode a broad variety of information, including all modalities of perceptual input as well as behavioral actions, and cognitive operations. In addition, hippocampal neurons seem especially tuned to relevant conjunctions among features that reflect unique episodic information, rather

148

Part II: Cognition

than simple perceptual or motor features. Moreover, hippocampal neural activity is particularly sensitive to modifications with experience such that alterations in the meaning of items or their relationships result in major changes in cellular firing patterns. Combined, these observations provide a bridge to current characterizations of declarative memory being accessible by many routes of expression, including literally one's verbal declarations about the contents of prior experiences. I begin with a conceptual scheme that could accomplish the coding of episodic information and the linking of this information into a larger network or memory space. Then I review the observations from studies on the firing patterns of hippocampal neurons that support this scheme. A scheme for hippocampal representation in declarative memory The evidence from both studies on human and animal amnesia and from brain imaging studies on humans points to an important role for the hippocampus in encoding complex information that contributes to our memory for personal experiences, that is, episodic memories, and for our ability to synthesize this episodic information into our body of world knowledge, fact or semantic memory. It is well known that the hippocampus receives information from virtually all sensory domains as well as other information about our internal states and our own behaviors (see Chapter 9). In addition, the hippocampus is remarkable for the extent of interconnectivity among the principal cells in its major processing areas. Also, the studies on hippocampal plasticity outlined in Chapter 3 have indicated that hippocampal neurons are particularly good at encoding conjunctions of diverse inputs. This combination of observations suggests a scheme in which subsets of hippocampal neurons could act as small networks for encoding episodic memories and for linking them together into larger networks that could support the properties of declarative memory. An outline of such a scheme is presented in Fig. 6-2. The scheme proposes three types or levels of coding by single hippocampal neurons. At the lowest level in this scheme, each hippocampal cell encodes a highly complex set of sensory and other information that composes a single behavioral "event." A single event is brief in its scope, akin to a single photographic snapshot that would include information about any salient stimuli, ongoing behavior, and the location or background in which the event occurred. In this scheme, single hippocampal neurons are viewed as capable of encoding the conjunction of all of this information that composes a single event captured in a brief moment of time.

Windows into the Workings of Memory

149

Figure 6-2. Schematic model of a simple "memory space." Events include ongoing behavior and location information, as indicated by individual numbered boxes. Episodes are composed of a sequence of event representations. The memory space is composed of a set of episodes linked by common, nodal events.

In addition, at the second level of this scheme, the activities of a set of hippocampal cells that are activated in sequence during a particular episode are preserved. This may be accomplished by cells whose plasticity properties can incorporate features of two or more sequential events, thus providing a mechanism for tying together the individual representations of multiple events. This combination of a set of sequential individual event codings and the codings that link multiple event sequences constitute the complete representation of a single episodic memory. Finally, at the third level of this scheme, it is proposed that there are other "nodal" hippocampal cells that encode features that are common across multiple episodes. According to this account, these cells receive strong inputs from some information about specific stimuli, behaviors, or location or other contextual information that is a salient element among otherwise distinct experiences. This feature could include a common critical stimulus to be remembered across different types of training trials, a common response one made to various stimuli in different trials, or a common location where different experiences occurred. The combination of these three functional prototypes of hippocampal cells provides the fundamental elements of a "memory space" that could support the properties of declarative memory. The event cells would cap-

150

Part II: Cognition

ture unique combinations of behavioral and location information that would mark specific episodes. The cells that encode sequences of multiple events could allow the recovery of entire episodes. The nodal cells could provide a bridge between episodic representations that could support the capacity for flexible and inferential memory expression by linking together indirectly information obtained across distinct episodes. What is the evidence that hippocampal cells have the properties of these prototypical units? A review of the story of efforts to characterize hippocampal neuron activity provides strong support for this model and insights into the workings of memory processing within the hippocampus. Early observations on the firing patterns of hippocampal neurons Following on the advent of technologies for recording the extracellular spike activity in behaving animals, several investigators began to explore the firing patterns of the large pyramidal cells in the hippocampus of rats. Electrophysiological techniques allowed these investigators to make the recordings in awake and behaving animals, providing the opportunity to correlate neuronal activations with stimulus events and motor patterns during a broad variety of behaviors, including learning. The expectations of investigators in these explorations were marked by caution. James Ranck, Jr., as he pioneered the earliest recording of hippocampal neurons in behaving animals, worried that cells in a brain structure located so many synapses from sensory input and motor output would have firing activity significant only as part of a large network; he suspected that neural firing patterns in response to external stimuli or behavioral output would be uninterpretable. But this clearly turned out not to be the case. Quite the opposite—hippocampal neuronal activity is well correlated with a very broad variety of stimuli and behavioral events, with the activity of cells "mirroring" virtually all the combinations of stimulus and behavioral events in any situation. Thus, identifying the scope and nature of information processing by hippocampal neurons has proved a formidable challenge, not because of the paucity of responses they might have evoked, but because of their variety, their complexity, and their plasticity in response to change. In Ranck's landmark 1973 paper, he described a large number of behavioral correlates of hippocampal neuronal activity. His categorization included cells that fired associated with specific orienting behaviors, approach movements, or cessation of movement, and with consummatory behaviors (feeding and drinking) or the mismatch of expected consum-

Windows into the Workings of Memory

151

matory event (e.g., the absence of water when it is usually found). At the same time, John O'Keefe was also recording from hippocampal cells in rats exploring open environment. His observations on hippocampal cellular activity led him to a different conclusion—rather than neuronal activity reflecting ongoing behaviors, O'Keefe and his colleague John Dostrovsky made the remarkable and historic discovery that hippocampal cells fire associated with locations the animal occupies at least as much as with the ongoing behavioral events. Ranck noted the possibility that his findings might be characterized in terms of the spatial specificity of firing patterns, and for some period the "behavioral" correlates of hippocampal firing patterns were largely overwhelmed with excitement about the spatial coding properties of these cells. This led to a period when considerable attention was given to the spatial properties of hippocampal neuronal activity, within the framework of the cognitive map theory of O'Keefe and Nadel, as introduced in the preceding chapter. What are place cells? Do they compose a "cognitive map"? The existence of location-specific neural activity in hippocampal neurons has been confirmed in many systematic studies. The common observation is that pyramidal neurons of the CA1 and CA3 fields of the hippocampus fire at high rates when the animal is in a particular location in the environment and fire little or not at all when the animal is located in other places. This observation led to the conclusion that each hippocampal cell has a distinct "place field," or area of the environment associated with high firing rate. Many current studies of place cells involve computerized tracking of the animal's position continuously in space, and automated means of determining the firing rate of the neuron associated with a matrix of locations. These studies show that many hippocampal cells have spatially specific activity that can be observed in many behavioral situations. A particularly clear example comes from a simple protocol where a rat forages for small food pellets distributed randomly throughout an open field. The rat continuously searches in all directions for extended periods (Fig. 6-3A). In this situation many hippocampal cells fire at a high rate only when the rat crosses a particular area in the environment (the place field), regardless of the rat's orientation within the environment. Once established, place fields can be very stable, and have been observed to show the same spatial firing pattern for months. However, the probability of firing of place cells is highly variable, in that sometimes the rate exceeds 100/sec on a pass through the field. Yet, on other passes, the cell may not fire at all, such that the average firing rate within the place field is typi-

Figure 6-3. Four different protocols used to map hippocampal place fields. For each protocol, the task is illustrated on the left. On the right is a mapping of an example place field (filled area), the location in the environment where the cell increased firing rate of the cell. Directional place fields are indicated by an arrow showing the direction of movement associated with increased firing. 52

Windows into the Workings of Memory

153

cally about 10/sec. This variability portends that factors other than the location of the animal per se determine the activity of these cells. In their initial report, O'Keefe and Dostrovsky realized immediately the potential significance of this neural correlate, and suggested that the hippocampus might subserve the creation and utilization of cognitive maps that animals use to navigate their environment, just as Tolman proposed in his efforts to characterize maze learning capacities in animals. In support of this view, O'Keefe later reviewed the existing findings on hippocampal place cells, focusing on the number and types of stimulus features in the environment that were encoded by place cells. He concluded that "a place cell is a cell which constructs the notion of a place in the environment by connecting together several multisensory inputs each of which can be perceived when the animal is in a particular part of the environment" (O'Keefe, 1979, p. 425). He supported this characterization by showing that the location-specific firing of most hippocampal neurons is controlled by the global configuration of the distant salient stimuli, and that any substantial subset of the spatial cues is sufficient to support spatially specific activity in some cells. More recent studies have shown that while some cells encode all of the available sensory cues, the locationspecific activity of most hippocampal neurons is controlled by a subset of the spatial stimuli. Therefore, the overall representation of space is a composite of many partial representations where each cell encodes the spatial relations between a few of the cues. The view that hippocampal cells encode spatial position is a critical feature of O'Keefe and Nadel's cognitive mapping hypothesis. The concept of hippocampal spatial representation that has emerged from these findings, illustrated in Fig. 6-4, is that the hippocampus contains a maplike representation of space. At a conceptual level, the map constitutes a coordinate grid, instantiated by intrinsic connections among hippocampal neurons. During investigation of a novel environment, representations of the relevant environmental stimuli are associated with appropriate spatial coordinate points. The resulting map is Cartesian in that it provides metric representations of distances and angles between the relevant stimuli. At the physiological level, a place cell reflects the occurrence of the rat at a particular coordinate position within the map. Thus, implicit in this model is the assumption that place fields can be considered "pointers" within a unified map, such that either every cell contains information about all of the cues or that cells representing subsets of the cues are all linked and bound by the global coordinate framework. O'Keefe and Nadel's central notion then is that the hippocampus constructs a facsimile of the environment, including the salient environmental cues.

154

Part II: Cognition

Figure 6-4. A schematic illustration of how an environment is mapped in the hippocampus, according to the cognitive mapping hypothesis (from Eichenbaum et al., 1999).

Despite the attractive features of a cognitive map in the hippocampus, there is little in the way of evidence that hippocampal cells act as elements of a cohesive map of the environment. Indeed, in contrast to this view, there is no topographical relationship between place cells in the hippocampus and locations in space. Furthermore, in many situations individual place cells are simultaneously controlled by different cues in the environment. So, while the existence of place cells as encoding locationspecific sensory information is now widely accepted, there is no strong evidence that these place representations are elements of a cohesive map of space. Instead, a strong possibility is that place cells represent familiar places as complex contextual information rather than as coordinates of a map. Additional evidence that place cell activity is strongly influenced by nonspatial factors suggests that location-specific activity may reflect the encoding of the places where important events happen. Hippocampal cells encode actions in places Do hippocampal place cells encode only the location of the animal? Even in his earliest description of place cells O'Keefe reported that the spatial activity of hippocampal neurons was influenced by more than just the location of the animal in the environment. Indeed, the preliminary study by O'Keefe and Dostrovsky emphasized that all the place cells fired only when

Windows into the Workings of Memory

155

the rat was facing a particular direction. O'Keefe's subsequent full analysis reported several variables in addition to location that determined place cell firing rate, including orientation, how long the rat was in the place field, and the elicitation of particular behaviors such as eating, grooming, and exploratory sniffing. Motion- and behavior-related correlates were the focus of Ranck's analysis of hippocampal firing properties, in which he described cells that fired primarily as a rat was involved in orientation, approach to particular objects or goals, consummatory movements, or cessation of movement. Several studies have now demonstrated that place cell firing is strongly influenced by movement direction and speed. Indeed, as rats perform in maze tasks, such as the radial maze, the majority of place cells in rats running on a radial maze fired almost entirely when the animal was running outward or returning inward on the maze arms (see Fig. 6-3B), and the firing rate was somewhat higher when the rat was running with greater speed. This finding is not dependent on the structure of the maze itself, but rather on the movement patterrns of the rat in the task. For example, in one study rats were required to begin each trial at the center of the open rectangular environment, and then could approach any of the four corners to obtain a reward. Subsequently they had to return to the center and then approach a different corner to obtain another reward, etc. The majority of place cells fired differentially according to the direction the rat was moving (see Fig. 6-3C). In addition, most place cells were also tuned for the speed of movement, and in many of these cells there was an optimal movement speed such that the cell fired at lower rates for both slower and faster movements through the place field. Also, the majority of the place cells fired differentially depending on the angle of the turn taken during movements. The activity of most cells was best described not so much in terms of spatial parameters but simply as firing at a particular time during the trip to or from a particular goal. A critical role for the animal's behavior, as much as the nature of cues available or the shape of the apparatus, was most directly demonstrated in a study that involved training rats in two different versions of the same task (see Fig. 6-3D). In one version the rats were initially trained to forage in an open environment for tiny food pellets distributed in random locations, causing the rats to move in all directions throughout the environment. In the second version of the task, the same rats were subsequently trained to find food only at four specific locations in the same open field that were repeatedly baited. Consistent with other descriptions of place cells of rats performing the random foraging task, in this situation the place cells were generally nondirectional, that is, fired similarly regardless

156

Part II: Cognition

of the direction of the animal's movement through the place field. By contrast, when the rats approached a small number of repeated reward locations, most of the cells (even the same cells) were directional, that is, fired only during the approach to, or the departure from, a particular location. These findings are distinct from the conventional characterization of place cells as localized with respect to static cues in the environment, and show that the ongoing behavior, defined as direction of movement, can strongly influence cell activity. Furthermore, these findings show that, even in the identical apparatus, place cells can exhibit a specific movement correlate depending on whether the task makes particular movement direction patterns significant to the structure of the task. Thus, in this context, the cells are not well characterized as place cells, because their activity does not reliably predict the animal's location. Rather, these cells seem to encode a particular relevant action in a particular place. This conclusion opens up the possibility that hippocampal cells may also encode other variables, including stimuli, behaviors, or other events that have no particular identity in location. Hippocampal neurons encode nonspatial stimuli and events As noted previously, even the earliest studies on hippocampal activity in animals exploring open environment included some cells that appeared to have nonspatial firing correlates. Consistent with these early findings, several investigators who have intentionally looked for event-related neural activity have demonstrated firing patterns of hippocampal neurons related to nonspatial stimuli and events. In most of these studies, as in the earliest place cell studies, the relevant stimuli and behaviors were confounded with spatial location in that each event is associated with the animal's presence in one place. Nevertheless, these studies show that nonspatial stimuli and events can be a necessary component of cellular activation, and a recent study provides compelling evidence of nonspatial firing patterns that occur independent of the animal's spatial location. Among the first findings of nonspatial correlates of hippocampal cellular activity were observations that hippocampal neurons fire associated with the development of a Pavlovian conditioned eyeblink, even when the animal was restrained throughout the training session. In addition, several studies have shown hippocampal neural activity associated with stimulus sampling in rats performing learned sensory discriminations or delayed matching or nonmatching to sample tasks. Notably in all these tasks other individual hippocampal cells exhibited striking responses associated with various behaviors including approach to the relevant discriminative stim-

Windows into the Workings of Memory

157

uli or the reward. Indeed, the activity of the population of hippocampal cells could be characterized as a set of neurons firing selectively at each phase of the task. Perhaps the most striking of these studies is a series of experiments by Sam Deadwyler and his colleagues, who have studied the firing properties of single hippocampal neurons and neuronal ensembles in rats performing delayed matching and nonmatching to sample in a discrete trials version of the task where the cues were left and right positions of two response bars. On each trial one bar was extended into the apparatus and pressed by the animal to obtain an initial reward. During the delay period the bar was withdrawn and the animal had to nose poke at a port on the opposite side of the apparatus. Finally both bars were extended into the apparatus and the animal could press the matching or nonmatching bar to obtain a second reward. In Deadwyler's analysis of single hippocampal cells, many of the neurons fired during the sample or match responses or upon the delivery of the reward, or combinations of these events. Many of these cells fired differentially depending on the left-right position of the bar pressed, or whether the response was correct or an error. In addition, many individual cells fired during the delay period, but their activity did not predict the position of the correct response. Deadwyler and colleagues' characterization of the activity of ensembles of cells recorded at several locations in the hippocampus focused on a statistical analysis that could extract patterns of covariances among the cells associated with different task events. These analyses showed that specific task parameters accounted for most of the overall variance in ensemble activity. The major components corresponded to encoding of the sample versus choice phase of the task regardless of bar position, encoding of the spatial position of the lever independent of task phase, encoding of left versus right error responses, and encoding of the sample position during the sample and choice phase. The encoding of lever position may be regarded as a spatial mapping. However, the coding of task phase cut across locations where the rat was positioned, and the coding of the sample position lingered into the choice phase when the rat was at the opposite bar on correct trials. Thus, while all the correlates of these cells can be considered "spatial" in various ways, a considerable amount of the variance in ensemble activity was not associated strictly with the animal's location but rather with the encoding of task-relevant spatial information. Analyses of the firing properties of hippocampal neurons have been extended to studies on the primate hippocampal region. In general the evidence for pure place-specific activity is poor, although sensory-evoked neural activity is often modulated by a variety of spatial variables. In humans, vi-

158

Part II: Cognition

sually evoked responses of hippocampal neurons have also been observed, and a substantial fraction of these cells fired on the sight of a particular word or face stimulus or during execution of task-relevant key press responses. In one recent study, Itzak Fried and his colleagues characterized the responses of hippocampal neurons in human subjects performing a recognition memory task with face and visual-object cues. Again a substantial number of cells responded to the stimuli, and individual cells had activity that differentiated faces from objects, or distinguished facial gender or expression, or new versus familiar faces and objects. The largest fraction of cells differentiated combinations of these features. Some of the cells had a specific pattern of responsiveness across all of these parameters. Some studies have been directly aimed at dissociating spatial and nonspatial firing patterns of hippocampal neurons by requiring animals to perform the same tasks using identical cues located at different locations in the environment. A particularly compelling study involved olfactory cues that were moved systematically among locations within a static environment, and provided unambiguous evidence of place-independent nonspatial hippocampal activity. Rats were trained to perform an odor-guided delayed nonmatch to sample task at multiple locations on a large open field (Fig. 6-5A). The stimuli were plastic cups that contained playground sand scented with one of nine common odors (e.g., coffee, cinnamon, etc.). On each trial one cup was placed randomly at any of nine locations on the open field. Whenever the odor differed from that on the previous trial (i.e., was a nonmatch) a Froot Loop was buried in the sand, and the rat would dig for the reward. Whenever the odor was the same as that on the previous trial (i.e., was a match) no Froot Loop was buried and the rat would turn away. The firing rate of hippocampal cells was assessed during the approach to the cups, focusing on the last second of the approach during which the animal arrived at the cup and generated its response. Firing rates were statistically compared across the set of odors, the set of locations, and match-nonmatch conditions. About two-thirds of hippocampal cells fired in association with one or more of these variables during the task. About one-third of the active cells' firing was not differentiated by the location of the cup, and about onethird of the active cells demonstrated some spatial component of firing. Some of the nonspatial cells were activated during the approach to any of the odor cups at any of the nine locations. Other cells fired differentially across the odor set (Fig. 6-5B), or between match and nonmatch conditions, or some combination of these variables and the approach. Only a small proportion of the location-selective cells fired associated only with the position of the cup (Fig. 6-5C). For the majority of cells, their activity was conjointly associated with the position where the cup was pre-

Windows into the Workings of Memory

159

Figure 6-5. Delayed nonmatching to sample test where the location of trials varies randomly. A: On those trials when the odor is different from that presented on the preceding trial (i.e., a nonmatch) the rat can dig in the scented food cup for a buried reward. On those trials when the odor is the same as that on the preceding trial (i.e., a match), there is no reward and the rat turns away. B: Example of a cell that fires associated with trials when odor 5 (O5) was presented but not when other odors were presented. The activity of this cell did not distinguish the locations where the trial was performed or the match-nonmatch status of the odor. C: Example of a cell that fired when trials were performed at adjacent locations P2-P3, but not when the trial was performed at other locations. The activity of this cell did not distinguish between different odors or their match-nonmatch status (data from Wood et al., 1999).

sented, the odor, the status of the odor as a match or nonmatch, and many of the cells fired at some point as the animal approached any cup. These results show that the activity of fully half of the activated cells acted completely independent of location, and most of the location-specific cells involved more than purely spatial features of the task. In addition, while

160

Part II: Cognition

some cells encoded particular odors or places, the activity of most cells was associated with one of the many potential conjunctions of odors, places, approach movements, and match-nonmatch events. These data indicate that, when important stimuli move unpredictably within an environment, a segment of the hippocampal population encodes the regularities of these stimuli without coding of the global topography. Combined with the other findings described before, these data provide compelling evidence that hippocampal cells can encode purely nonspatial information that is relevant at many locations in the environment. The hippocampal network mediates a "memory space" The preceding review both confirms the existence of location-specific activity of hippocampal neurons, and suggests that place cells are parts of a neural representation that is both less than, and more than, a map of space. The representation is less than a map because it is clear that place cells individually and independently encode pieces and patches of an environment without acting as part of a full mapping. The representation is also more than a map because spatial codings are modulated by relevant nonspatial task variables and hippocampal cells encode nonspatial stimuli and events. How can the impressive location-specific activity be reconciled within a larger framework of memory representation? The properties of hippocampal firing patterns are entirely consistent with the memory space scheme proposed earlier in this chapter. To place the properties of hippocampal firing patterns in the context of this scheme, let us review the firing patterns of hippocampal cells in rats performing two prototypical tasks, a spatial task similar to the radial maze task, and an odor-guided memory task. Begin with the data from rats performing the spatial memory task in a large arena (see Fig. 6-3C). Many of the cells can be described as place cells having a place field in a particular portion of the arena, with their activity modulated by several movement parameters such that the cells fire maximally when the rat is moving at a particular speed, in a particular direction, and when the rat is turning in one particular angle. One cell, whose firing patterns are depicted in Fig. 6-3 C, can be described as a cell with a place field near the center of the arena, and its activity is modulated by several movement parameters: It fires maximally when the rat is moving at a moderate speed, when the rat is moving "north," and when the rat is turning slightly to the right. Alternatively, however, virtually all of the activity of this cell can be characterized more simply as firing when the rat initiates its approach to cup 1. Most cells recorded as rats performed this task could

Windows into the Workings of Memory

161

similarly be described by a complex combination of place, direction, and turning angle, or could be at least as well described by when the cell fired during movements toward, or in return from, a particular reward cup. Furthermore, the population of hippocampal neurons contains cells that fire at virtually every point in the path to and from each of the cups (Fig. 6-6A). So one can best describe the full data set as a network of neurons each of which encodes one fragment of the approach and return from each reward cup. Of course, within the memory space framework outlined previously, the hippocampal network representation contains a subset of cells that represents distinct events, each defined by a specific location and a specific set of movement attributes, linked to encode a particular kind of trial episode, defined as the approach to and return from a particular goal location. Now let us expand the same kind of characterization to hippocampal cellular activity in rats performing the odor-guided tasks such as those described earlier, with the additional insight into how "secondary" nonspatial firing properties are accommodated into hippocampal spatial activity (Fig. 6-6B). As animals perform these tasks individual cells activate at vir-

Figure 6-6. Temporal organization of hippocampal neuronal activity. A: Schematic illustration of firing patterns of a set of cells as the rat performs a spatial memory test. Different cells fire at each successive moment as the rat approaches each cup and returns. Note that each cell has a place field and directional firing preference associated with a particular segment of a particular outward-bound or inward-bound episode. B: Schematic illustration of firing patterns of a set of cells as the rat performs an odor discrimination task. Different cells fire at each successive moment as the rat approaches the odor ports, samples the odors, and retrieves the reward, and other behaviors. Some cells fire on all trials, others fire only if a particular odor is presented (data from Eichenbaum et al., 1999).

162

Part II: Cognition

tually every instant of the task, defined as time-locked neural activity to identifiable stimulus and behavioral events. Individual cells in animals performing the odor-guided nonmatch to sample task fired at different times associated with the approach to the odor sampling port, during the odor sampling period, and during the discriminative response and reward period. Thus, overall hippocampal activity in this situation, like that in the spatial task described before, can be described as a network of cells that encode each event that characterizes each type of discrimination trial episode. In addition, in both spatial and odor tasks, some of the hippocampal cells showed considerable specificity for particular locations of approach or odors sampled, whereas other cells fired on every trial for different directions of movement and different types of odor trials. This observation leads to an additional characterization of how hippocampal cells are "tuned" to brief segments of behavioral episodes. Let us first consider the cells that show the greatest specificity for particular conjunctions or relations among stimuli or events. Prominent examples include conjunctions or relations between the places and actions that occur in those places, or between sequences of odors. A similar characterization can be offered to describe the data from the odor-guided delayed nonmatch to sample task where the trials were performed at different locations on an open field. Some cells associated only with particular conjunctions of events, places, or both, and virtually every possible conjunction was represented. These most selective cells might be thought of as reflecting ever more rare variations of events. Now let us consider the cells that fired consistently on different types of trials in spatial and odor tasks. Indeed, many of these cells fired associated with locations or events that were common across all the different types of trials within a task. In the spatial task, some cells fired whenever the animal's path crossed a particular place regardless of what cup was being approached or left behind. In the odor tasks, some cells fired during a particular common behavioral event, for example, during the approach movement regardless of spatial location, and others fired on all match or nonmatch trials. In the task where the odor test was performed at multiple locations some cells fired on all trials at a particular location regardless of the current odor or type of trial. These cells might be thought of as the "nodal" cells introduced in the scheme presented in Figure 6-2, the cells that encode intersections among the episodes that have in common a particular place or a particular stimulus or behavior. Of course, the full range of cells extends continuously from the most common nodal point cells to the most highly specific cells. From this perspective, "pure" place

Windows into the Workings of Memory

163

cells (ones that are location specific and have no other quality) are one of the more common nodal correlates, whereas the highly combinatorial cells represent events that occurred only a few times in a session. Evidence of episodic-like representations in the hippocampus More direct evidence of coding for information specific to particular types of episodes comes from another experiment where hippocampal cells fire differentially even in situations where the overt behavioral events and the locations in which they occur are identical between multiple types of experience. In this experiment rats performed a spatial alternation task, a simple version of one of Olton's episodic memory tasks, performed in a T-maze. Each trial commenced when the rat traversed the stem of the T and then selected either the left- or the right-choice arm (Fig. 6-7A). To alternate successfully the rats were required to distinguish between their left-turn and right-turn experiences and to use their memory for the most recent previous experience to guide the current choice. Different hippocampal cells fired as the rats passed through the sequence of locations within the maze during each trial. Most important, the firing patterns of many of the cells depended on whether the rat was in the midst of a leftor right-turn episode, even when the rat was on the stem of the T and running similarly on both types of trials—minor variations in the animal's speed, direction of movement, or position within areas on the stem did not account for the different firing patterns on left-turn and right-turn trials (Fig. 6-7B). Other cells fired when the rat was at the same point in the stem on either trial type. Thus, the hippocampus encoded both the leftturn and right-turn experiences using distinct representations, and included elements that could link them by their common features. In each of these experiments, the representations of event sequences, linked by codings of their common events and places, could constitute the substrate of a network of episodic memories. Elaboration of a general scheme for memory representation by the hippocampus There are three central aspects of this novel characterization of the firing properties of hippocampal neurons in animals performing a broad range of learning and memory tasks. First, cellular activity can be described as a sequence of temporally and spatially defined events that constitute each trial. Second, some cells show a very high degree of specialization, such as the approach to a particular odor at a particular place only when it is

164

Part II: Cognition

Figure 6-7. T-maze alternation. A: Illustration of trials in this task. On each trial the rat must remember the previous episode and then turn in the opposite direction. B: Example of a hippocampal cell that was active when the rat is traversing the stem section of a T-maze while performing the spatial alternation task. This cell fired almost exclusively during left-turn trials. In the left and middle panels, the paths taken by the animals on the central stem are plotted for left-turn trials (light gray) and for rightturn trials (dark gray). The locations of the rat when individual spikes occurred are indicated by dots for left-turn trials (left panel), and then right-turn trials (middle panel). In the right panel, the mean firing rate of the cell for each of four sectors of the maze, adjusted for variations in firing associated with other behavioral factors, is shown separately for left-turn trials (left bars) and right-turn trials (right bars). (significant differences: ** p